Department of Business administration

BUSN89

Degree Project in Corporate Financial Management – Master level

Spring 2014

The Accruals Based Trading Strategy on

the Swedish Stock Market

Does the benchmark when classifying extreme accrual firms have an impact on

the trading strategy’s effectiveness?

Authors:

Jessica Buö

Ellinor Molander Kroon

Supervisor:

Maria Gårdängen

2

Abstract

Title: The Accruals Based Trading Strategy on the Swedish

Stock Market

Seminar date: 2014-06-02

Course: BUSN89 Degree Project in Corporate Financial

Management - Master Level

Authors: Jessica Buö, Ellinor Molander Kroon

Supervisor: Maria Gårdängen

Key words: Accruals anomaly, trading strategy, market efficiency,

earnings fixation

Purpose: To investigate if it is possible to earn abnormal returns

from the accruals based trading strategy in Sweden. The

aim is also to examine if the benchmark used when

classifying firms into accruals portfolios has an impact

on the abnormal returns from the trading strategy.

Methodology: The study investigates the abnormal returns from the

trading strategy that aims to exploit the accruals

anomaly. Firms are divided into portfolios and using

previously established methods of risk-adjustment, the

abnormal returns from the trading strategy are estimated

and tested for each of the 11 portfolio years.

Theoretical perspectives: The theory in the paper is based on previous research on

the accruals anomaly. The theories have been applied

when investigating other markets to describe and

analyse the accruals anomaly.

Empirical foundation: The study examines firms listed on NASDAQ OMX

Stockholm main market. Portfolios are formed each year

between 2002 and 2012. The majority of the data has

been collected using Thomson Reuters Datastream.

Conclusions: The results show that while the benchmarks classify

firms differently, the difference does not spill over to the

abnormal returns earned by the investor. Overall, the

results for all risk-adjustment methods show that

investor may not earn positive abnormal returns from

the trading strategy in Sweden.

3

Contents

1. INTRODUCTION .................................................................................................................. 5

1.1 Background ....................................................................................................................... 5

1.2 Problem discussion ........................................................................................................... 6

1.3 Purpose .............................................................................................................................. 8

1.4 Research question ............................................................................................................. 8

1.5 Delimitations ..................................................................................................................... 8

1.6 Outline ............................................................................................................................... 9

2. THEORETICAL FRAMEWORK ....................................................................................... 10

2.1 Theories on market efficiency ........................................................................................ 10

2.1.1 The efficient market hypothesis ............................................................................... 10

2.1.2 Market inefficiencies ................................................................................................ 10

2.1.3 The accruals anomaly ............................................................................................... 11

2.2 Theories on earnings and its importance to investors ..................................................... 11

2.3 Theories on the accruals anomaly ................................................................................... 12

2.3.1 Earnings fixation hypothesis .................................................................................... 12

2.3.2 The agency theory of overvalued equity .................................................................. 13

2.3.3 Alternative findings on the accruals anomaly .......................................................... 14

2.3.4 Accounting and institutional setting as contributors to the accruals anomaly ......... 15

2.3.5 Theories contradicting the existence of the accruals anomaly ................................. 16

2.3.6 The accruals anomaly in Sweden ............................................................................. 17

3. METHODOLOGY ............................................................................................................... 19

3.1 Data and sample selection ............................................................................................... 19

3.1.1 Biases in accrual research ......................................................................................... 20

3.2 Variables ......................................................................................................................... 20

3.2.1 Accruals .................................................................................................................... 21

3.2.2 Earnings .................................................................................................................... 22

3.2.3 Cash flows from operations ...................................................................................... 22

3.3 Classifying firms according to their accruals using different benchmarks ..................... 23

3.4 Returns ............................................................................................................................ 24

3.4.1 Stock returns ............................................................................................................. 24

3.4.2 Abnormal stock returns ............................................................................................ 25

3.4.3 Buy-and-hold abnormal returns using reference portfolios ...................................... 26

3.4.4 Calendar time approach ............................................................................................ 29

4

3.4.5 Winsorization............................................................................................................ 33

4. RESULTS AND ANALYSIS .............................................................................................. 34

4.1 Comparing the benchmarks ............................................................................................ 34

4.1.2 Market capitalization ................................................................................................ 34

4.1.3 Stock price ................................................................................................................ 35

4.1.4 Net income and Cash flows from operations ............................................................ 36

4.2. Results from testing the existence of the accruals anomaly in Sweden ......................... 37

4.2.1 Total assets benchmark ............................................................................................. 38

4.2.2 Earnings benchmark ................................................................................................. 39

4.2.3 Comparing the results to previous research .............................................................. 41

4.2.4 Results from testing the trading strategy after winsorizing outliers ......................... 48

5. CONCLUSION .................................................................................................................... 52

REFERENCES ......................................................................................................................... 54

Academic literature ............................................................................................................... 54

Books .................................................................................................................................... 56

Student work ......................................................................................................................... 56

Additional sources ................................................................................................................ 56

APPENDIX .............................................................................................................................. 57

Appendix 1 ............................................................................................................................ 57

Appendix 2 ............................................................................................................................ 58

Appendix 3. ........................................................................................................................... 59

5

1. INTRODUCTION

The first chapter provides a background and discussion on the subject of this paper before

presenting the chosen research questions. The purpose of the study is defined as well as any

delimitations.

1.1 Background

In 1996, Richard Sloan (1996) presented a paper that suggested that there is a possibility to

earn abnormal returns on the US market by following a trading strategy based on the

components of a firm's reported earnings. The reasoning behind this was that firms’ earnings

can be divided into a cash flow component and an accruals component. The cash flow

component represents an actual cash transaction, while the accruals component adjusts for a

mismatch between cash transactions and the actual delivery of goods or services1. When

Sloan published the results from his study, he showed that accruals are less persistent in future

earnings than cash flows, but that investors fail to recognize this. Instead, investors tended to

overweight the value of current accruals and underweight the value of current cash flows

when they forecasted future earnings. This has become known as the accruals anomaly and

represents a systematic mispricing on the market that allows for the possibility to earn

abnormal returns on stock investments. Sloan (1996) constructed a trading strategy designed

to take advantage of the situation, where firms were sorted into portfolios based on the

relative size of their accruals. The reasoning behind the strategy was that when accruals were

reversed in the future periods, investors were surprised by the reported earnings, which was

reflected in the stock price. Hence, firms that had reported high accruals experienced a falling

stock price as earnings did not meet expectations due to the reversal of accruals. Thus, by

taking a long position in firms with low accruals and a short position in firms with high

accruals, Sloan (1996) found that the trading strategy was able to attain double-digit abnormal

returns on the US market. The findings received widespread recognition in academic

literature (Green et al., 2011). In fact, they were taken beyond the scope of the academic

world and used by professionals in investment decisions, as Business Week reported in its

cover story on 4 October, 2004:

1 Examples of accruals are accounts payable or accounts receivable, which may impact a firm’s earnings, but do not correspond to a similar

change in cash flows.

6

“Now, Goldman Sachs Asset Management, Barclays Global Investors, and Susquehanna

Financial Group, among others, are employing versions of the Sloan-Richardson Models to

guide their investments.” (Henry, 2004, p. 79-88)

Furthermore, the findings challenge one of most widely accepted theories on capital markets;

the efficient market hypothesis (Hafzalla et al., 2007). The efficient market hypothesis was

introduced in 1970 by Eugene Fama, who was recently awarded Sveriges Riksbanks Prize in

Economic Sciences in Memory of Alfred Nobel for his contributions on the matter. (Nobel

Foundation, 2013) The efficient market hypothesis suggests that all information available on a

firm should be reflected in its stock price. As reported earnings and its components are

publicly available information, the market should incorporate the information into the stock

price accordingly. (Fama, 1970) However, the magnitude and duration of abnormal returns

found in Sloan (1996) is an indication that the market efficiency fails to hold and has become

a highly debated issue in finance and accounting literature (Hafzalla et al., 2007).

1.2 Problem discussion

While a majority of the studies on the US stock market have been fairly unanimous regarding

the existence of the accruals anomaly (see for example Sloan, 1996, Collins & Hribar 2000;

Bradshaw et al., 2001; Xie, 2001; Zach, 2003; LaFond, 2005; Hafzalla et al., 2007; Pincus et

al., 2007), there are researchers that have found otherwise. For example, Kraft et al. (2006)

point out that results in previous research are rarely robustness checked by controlling for

outliers and look-ahead bias (discussed in detail in chapter 3), and argue that when these

biases are accounted for, the anomaly is no longer present. This suggests that the results found

in previous studies are due to a few, extreme observations, and should not be generalized to a

greater population. Furthermore, Khan (2008) and Wu et al. (2009) argue that the accruals

anomaly is largely attributable to misspecified risk adjustments, suggesting that the findings

are sensitive to the chosen asset pricing model used to specify expected returns. Finally,

Green et al. (2011) and Johnson & Schwartz (2001) found evidence that the anomaly has

declined over the years. Green et al. (2011) argue that the increased research on the subject

has inclined investors to exploit the anomaly, leading to the abnormal returns from the trading

strategy being arbitraged away. This argument has also been backed by practitioners, for

7

example Ron Kahn, Global Head of Equity Research at Barclay’s Global Investors, who was

quoted in Financial Times:

“Buying companies with high quality earnings and shorting those most dependent on

accruals proved a good strategy, until the market figured it out.” (Skypala, 2009, p.4)

In spite of the great amount of attention the subject has received, the majority of research to

date has been conducted on the US market while few studies have been done on the Swedish

market. In addition, the research that has been conducted, has reached ambiguous results

regarding the existence of the accruals anomaly in Sweden. For example, LaFond (2005)

found that the accruals based trading strategy in Sweden does yield positive abnormal returns,

but only at a 10% significance level. Pincus et al. (2007) on the other hand, found no

statistical support for this. Furthermore, one recent master thesis (Giaever & Gabrielsson,

2007) supports existence of the accruals anomaly in Sweden, while another (Ivanov &

Maximova, 2011) fails to find such evidence. The ambiguity in results on the Swedish market

suggests the need for further research in order to determine if the anomaly is present in

Sweden, in which case investors may exploit it and earn abnormal returns on their investment

portfolio.

One factor that all studies on the Swedish market have in common is that they benchmark the

accruals by total assets when classifying firms into accrual portfolios in the trading strategy.

By setting the size of accruals in relation to the size of the firm, the researchers enable for

comparison between firms, which is necessary when comparing portfolio returns (Sloan,

1996). However, Hafzalla et al. (2007) suggest a new benchmark for classifying the extreme

accrual firms, namely by earnings. Their results suggest that by modifying the classification

measure in this way, the significance of statistical results improve, thus making the trading

strategy more effective. In an attempt to contribute to the limited research and ambiguous

results on this topic in Sweden, this paper will apply Hafzalla et al.’s (2007) classification

method, as well as Sloan’s (1996). The aim is to investigate if the new benchmark will detect

higher statistically significant results of the accruals based trading strategy on the Swedish

market than the classification using total assets.

8

Finally, due to the criticism by Khan (2008) and Wu et al. (2009) against previous literature’s

methodology, this study will employ several measures of risk-adjustments. This is further

motivated by the difference in results across studies on the Swedish market, which suggests

that there may be a misspecification in estimating abnormal returns. This study therefore

includes the Capital Asset Pricing Model, Fama-French 3 factor model and buy-and-hold

abnormal returns using size-matched reference portfolios. A distinction will also be made

between earlier and later years in order to investigate if the accruals anomaly has been

arbitraged away in recent years, as is argued by Green et al. (2011) and Johnsson & Schwartz

(2001).

1.3 Purpose

As the benchmark used to identify extreme accrual firms changes, the effectiveness of the

accruals based trading strategy has been found to improve accordingly (Hafzalla et al., 2007).

The aim of this study is to investigate if the recently introduced earnings benchmark by

Hafzalla et al. (2007) can improve previous researchers’ low significance of an accruals based

trading strategy on the Swedish stock market. If the accruals anomaly is present on the

Swedish market, investors may use this information in order to benefit from abnormal returns.

1.4 Research question

There are two main questions that this paper will research:

1. When attempting to create an accruals based trading strategy, to what extent is the measure

that classifies accruals by earnings more effective than the measure that classifies accruals by

total assets?

2. To what extent is it possible to earn abnormal returns from the accruals based trading

strategy in Sweden?

1.5 Delimitations

The scope of this study is limited to only include non-financial firms listed on NASDAQ

OMX Stockholm main market. This is due to the special nature of the financial statements of

financial firms and to increase comparability with previous research that frequently excludes

such firms.

9

1.6 Outline

Chapter 2 starts with highlighting theories and previous research that is relevant to the chosen

research questions. Chapter 3 describes the methodology when classifying firms into

portfolios and estimating abnormal returns of the trading strategy. Chapter 4 presents the

results and analysis with regards to theory. Chapter 5 presents conclusions and proposes

topics for future research.

10

2. THEORETICAL FRAMEWORK

The second chapter introduces the reader to theories related to the accruals anomaly and

relevant research. The chapter is split into three main sections. The first is the debate on

market efficiency and the existence of anomalies. The second discusses earnings components,

and their relation to future performance. Finally, the third section presents arguments on why

the accruals anomaly may or may not exist.

2.1 Theories on market efficiency

2.1.1 The efficient market hypothesis

The efficient market hypothesis, presented by Eugene Fama in 1970, predicts efficient

financial markets where all information available on a firm is reflected in the stock price.

Thus, it is not possible for investors to exploit mispriced stocks and make arbitrage profits

since the stock price already accurately incorporates all information about the firm. According

to Fama (1970) there are three levels of market efficiency. A market that has weak efficiency

only reflects historical information in stock prices, making it difficult to make arbitrage

profits through a technical analysis. A semi-strong market is extended to include current

public information, and it is therefore not possible to arbitrage from fundamental analysis.

The third form, strong efficiency, fulfils both criteria above but in addition assumes that all

insider information is also available to the public. This state is generally not considered to be

realistic and instead the semi-strong state is the most accepted. (Fama, 1970)

2.1.2 Market inefficiencies

Market anomalies are circumstances of market inefficiency where stock prices are

systematically mispriced and do not reflect all public information available. (Shefrin, 2007)

There are two views on market anomalies. The first is the traditionalists’ view, which claims

that market anomalies are small and temporary. The second is the behaviourists’ view, which

argues that anomalies can be large and more permanent. Thus, according to both views, it

may be possible to exploit anomalies in order to earn abnormal returns; however they differ

with regards to the time frame and magnitude. (Shefrin, 2007) Still, Shefrin (2007) is clear to

point out, that neither view is in support of the possibility of permanent market inefficiency.

With respect to the accruals anomaly, this view opposes that proposed in previous literature

(such as Hafzalla et al., 2007) which often argues that the existence of the accruals anomaly is

evidence against market efficiency.

11

While certain anomalies are linked to time effects or structural factors, Shefrin (2007)

attributes investors’ failure to price stocks according to a firm’s fundamental value to

behavioural phenomena. As investors may view risk and return differently depending on the

circumstances and market conditions, they are not always rational when pricing stocks which

can lead to systematic mispricing. (Shefrin, 2007)

2.1.3 The accruals anomaly

Sloan (1996) found that investors fail to price stocks correctly with respect to firms’ reported

accruals, i.e. the part of earnings that is not attributed to cash flows from operating

activities. By investigating the components of firms’ earnings and go long in firms with high

accruals and short low accrual firms, Sloan concluded that investors were able to earn

abnormal returns of 10.4%. (Sloan, 1996) This occurrence of a systematic mispricing is

referred as the accruals anomaly and further research on the subject will be presented in the

section 2.3.

2.2 Theories on earnings and its importance to investors

Ball and Brown (1968) suggested that there is a positive relationship between earnings and

stock returns. This is due to the fact that earnings are considered to summarize value relevant

information about a firm. By breaking down current earnings into its components, one may

investigate a firm’s future earnings power, defined as “the level of earnings an enterprise can

be expected to sustain over the next five to ten years.” (Graham, et al. 1962, referred to by

Sloan, 1996, p. 290) The argument is that certain components of earnings are more persistent

when forecasting earnings and are thus more reliable when assessing the firm’s future

earnings power. (Graham, et al. 1962, referred to by Sloan, 1996) In particular, several

authors (Sloan, 1996; Collins & Hribar, 2000; Bradshaw et al., 2001; Xie, 2001; Zach, 2003;

LaFond, 2005) have found that operating accruals are less persistent in future earnings than

operating cash flows. Forecasts should therefore be adjusted for the accruals, in order to avoid

a future earnings surprise. (Graham, et al. 1962, referred to by Sloan, 1996) Furthermore,

Bernstein (1993) pointed out that cash flows from operations are less exposed to subjective

valuations than other components of earnings, such as accruals. Therefore, a higher ratio

between cash flows from operations and net income suggests a higher earnings quality. In the

meantime, the opposite may suggest that the firm’s management is involved in earnings

management activities such as using doubtful accrual criteria. (Bernstein, 1993) In spite of

12

this, several researchers have found that investors fail to evaluate the different components of

earnings when forecasting future performance. (Ou & Penman, 1989; Bernard & Thomas,

1990; Hand, 1990; Maines & Hand, 1996; Sloan, 1996; Dechow et al., 2005) As a result, the

positive relationship between earnings and stock returns is likely due to investors’ fixation on

earnings when pricing a stock, rather than earnings ability to summarize information on firm

value.

The relevance of market efficiency and earnings to this study

Since financial markets today are assumed to be semi-strong and accounting data are

available to the public, investors should not be able to earn abnormal returns by trading on

this kind of information. In other words, in order for the market efficiency to hold, one should

not be able to earn abnormal returns on an accruals based trading strategy. If the results in this

study show that is possible, it suggests that the Swedish market fails to recognize the

information available on accruals when assessing the quality of a firm’s current earnings and

its future earnings power.

2.3 Theories on the accruals anomaly

2.3.1 Earnings fixation hypothesis

When Sloan (1996) studied US firms between 1962 and 1991, he found that cash flows are

more persistent than accruals in earnings, but that the market fails to anticipate this when

forecasting future earnings. Due to this earnings fixation, investors are surprised when

accruals are reversed in subsequent periods, and returns alter to reflect the stock price changes

caused by the earnings surprise. (Sloan, 1996) For example, firms that recorded high earnings

due to high accruals may initially have generated positive returns but experienced a falling

stock price once the accruals were found to be overstated or reversed. Sloan found that this

systematic mispricing may be exploited and developed a trading strategy accordingly. He

sorted firms into portfolios based on the size of their accruals, benchmarked by total assets.

The abnormal return in the lowest accrual portfolio was 4.9% and –5.5% in the highest

accrual portfolio. He then created a hedge portfolio where he took a long position in firms that

reported low accruals and a short position in firms with high accruals. By doing this, Sloan

found that an investor could earn abnormal returns of 10.4%. (Sloan, 1996) Since then, a

number of studies have confirmed Sloan’s work and the existence of the accruals anomaly,

13

both in the US as well as in other markets. (Collins & Hribar, 2000; Bradshaw et al., 2001;

Xie 2001; Zach, 2003; LaFond, 2005; Pincus et al., 2007)

A decade after Sloan published his paper, Hafzalla et al. (2007) investigated the accruals

anomaly on US market between 1988 and 2004. In contrast to most previous research, they

applied a new benchmark for identifying extreme accrual firms, namely by classifying

accruals in relation to earnings. This differed from Sloan (1996) and the majority of earlier

research who applied total assets as a benchmark to allow for comparison between firms.

Hafzalla et al. (2007) found that by modifying the classification measure to instead scale by

earnings, the significance of their statistical results for the trading strategy improved, thus

making it more effective. Using the total assets benchmark, they found an abnormal return of

-6.3% for highest accrual portfolio and 2.6% for lowest accrual portfolio. The latter was

however, not statistically significant. The abnormal return for the entire hedge portfolio was

9% but statistically insignificant. When using the earnings benchmark, the abnormal return

was -6.9% for highest accrual portfolio, 5.4% for lowest accrual portfolio and 12.3% for the

entire hedge portfolio. All results were significant at the 1% level. In addition to the statistical

improvement, Hafzalla et al. (2007) argued that by dividing accruals by earnings instead of

total assets, the interpretation became more intuitive when relating to Sloan’s (1996) earnings

fixation hypothesis. This was due to the fact that the classification became a matter of

accruals as a proportion of earnings, rather than accruals representing a growth of total assets.

(Hafzalla et al., 2007)

2.3.2 The agency theory of overvalued equity

Kothari et al. (2006) suggest an alternative explanation to the accruals anomaly. The authors

found that companies in the highest accrual portfolio experienced large abnormal returns prior

to the event year (the year where they were defined to have high accruals) which was then

followed by stock underperformance in the subsequent year. However, the relationship was

asymmetric, as high accrual firms’ returns were found to be -7.6% and statistically significant

but low accruals firms’ returns were insignificant, (-1.92%). The authors argued that this

asymmetry cannot be explained by the earnings fixation hypothesis, and supported the agency

theory of overvalued equity argument (Jensen, 2005) instead. The agency theory of

overvalued equity predicts that managers of overvalued firms are more likely to manipulate

their earnings and adjust accruals upwards in an attempt to prolong the overvaluation.

14

Undervalued firms on the other hand, do not have the same incentive to manipulate earnings

downwards. This resulted in overvalued firms frequently appearing in the high accrual

portfolio, but undervalued firms being more spread out, and not concentrated to the low

accrual portfolio. (Kothari et al., 2006)

The relevance of the earnings fixation hypothesis and agency theory of overvalued equity to

this study

Based on Hafzalla et al.’s (2007) findings is interesting to investigate whether the

classification based on accruals scaled by average total assets (Sloan, 1996) or scaled by

earnings (Hafzalla et al., 2007) generates the most profitable hedge portfolio. If this study

finds that it is possible to earn abnormal returns from the accruals based trading strategy in

Sweden, the results will be compared to previous research to see which theory the results

support. A significant negative relationship between Swedish firms’ reported accruals and

stock returns, suggest that the results are consistent with the earnings fixation hypothesis.

However, if the study finds statistical significance only for high accrual firms, and they

experience negative returns, the earnings fixation hypothesis is not applicable on the Swedish

market. In that case, the results instead support the agency theory of overvalued equity. The

findings may be interpreted as evidence that the accruals anomaly on the Swedish market is a

result of managers’ attempts to prolong firms’ overvaluation. Thus, the overvalued equity

argument differs from the earnings fixation hypothesis as it assumes that the accruals anomaly

is a consequence of mispricing rather than the cause itself.

2.3.3 Alternative findings on the accruals anomaly

Kraft et al. (2006) researched the existence of the accruals anomaly on NYSE /AMEX firms

between 1987 and 1999. Their results indicated an inverted U-shape in the relationship

between abnormal returns and firm accruals, in other words, they found significantly negative

abnormal returns for both the extreme accrual portfolios. They employed a trimming

procedure of 1% that eliminated a small number of outlying observations in order to

robustness test their findings. Kraft et al. (2006) argued that the inverted U-shape in abnormal

returns could not be explained by Sloan’s (1996) earnings fixation hypothesis as earnings

fixation does not predict the significantly negative returns in the low accrual portfolio. A main

argument proposed in the paper was that of the lack of robustness tests in previous research

when assessing market mispricing. The authors argued that when disregarding the importance

15

of robustness testing, findings may easily be misleading as outliers cause misspecified test

statistics.

2.3.4 Accounting and institutional setting as contributors to the accruals anomaly

In contrast to many previous studies that have only focused on the US market, Pincus et al.

(2007) investigated the international presence of the accruals anomaly. They specifically

investigated how accounting and institutional factors added to the anomaly’s existence, and

found that legal tradition, allowance for accruals accounting, ownership concentration and

limits to insider-trading all have significant impact. Firstly, Pincus et al. (2007) found that the

anomaly is more likely to occur in common law countries. Their findings were unexpected as

countries with a common law tradition are generally associated with stronger investor rights

and greater access to firm specific information (LaPorta et al., 1998). Furthermore, previous

research by Hung (2001) found that financial markets in common law countries are assumed

to have a greater degree of efficiency, suggesting that accrual mispricing would be less likely.

In spite of this, the results of Pincus et al. (2007) indicated that even though investors were

assumed to have greater access to information, this was the legal setting where the accruals

anomaly was mostly likely to appear. Pincus et al. (2007) explained the anomaly’s occurrence

by claiming that investors in common law countries seem to disregard their access to

information and focused more on total earnings rather than its underlying components. The

authors also pointed out that extensive use of accrual accounting in certain countries could

add to the anomaly’s existence. As accruals are not always explicitly reported in financial

statements, Pincus et al. (2007) argued that there is an information gap which increases the

incentives for managers to adjust the earnings in a certain matter. Finally, Pincus et al. (2007)

concluded that restrictions on insider trading could affect the duration of the anomaly’s

existence. In countries that had a more strict attitude towards insider-trading, insiders could

not trade based on their knowledge of the persistence of the firm’s accruals. As a result, the

accruals mispricing on the market was not arbitraged away as quickly.

The relevance of accounting and institutional setting to this study

As Sweden is considered a civil law country (La Porta et al., 1998), the findings of Pincus et

al. (2007) would suggest that the accruals anomaly will likely not appear here. However, as

will be discussed in the section 2.3.6 below, LaFond (2005) found significant abnormal

returns from the accruals based trading strategy in Sweden. This suggests that if the results of

16

this study show that abnormal returns may be achieved from the trading strategy, Sweden

may be an exception to the findings of Pincus et al. (2007)

2.3.5 Theories contradicting the existence of the accruals anomaly

If it is well known that it is possible to earn abnormal returns by looking at the accruals

component in firms’ earnings, should not the market exploit this and, after a while, the

mispricing be arbitraged away? When Green et al. (2011) investigated the accruals anomaly

on the US market between 1989 and 2008, they found that after 2004, the positive raw and

risk-adjusted returns based on Sloan’s (1996) trading strategy seemed to have declined. As

this was compelling evidence against many earlier studies, they evaluated their results

carefully with several specification and robustness tests. They also tested the implication of

using different accruals definitions and measurement techniques. The findings were robust

and they concluded that abnormal returns due to an accruals based trading strategy were no

longer positive, indicating that the market had recognized the mispricing and arbitraged it

away. The authors explained the disappearance of the accruals anomaly to be partly due to

that many hedge funds have hired accounting scholars and invested a lot of capital in extreme

accrual firms. By doing this, Green et al. (2011) argued that accounting researchers may have

identified a mispricing on financial markets but that they also helped practitioners exploit

them to the point they no longer existed.

As mentioned in the introductory chapter, there are other researchers that also have found

evidence that the accruals anomaly may not be the phenomenon that some researchers claim.

Khan (2008) and Wu et al. (2009) for example argue that the accruals anomaly is largely

attributable to a misspecified risk adjustment, suggesting that the findings are sensitive to the

chosen asset pricing model used to specify expected return. This is will discussed further in

chapter 3, but motivates the use of different methods of risk adjustment in order for the results

to be considered reliable.

The relevance of theories contradicting the existence of the accruals anomaly to this study

If the results of this study show that the abnormal returns on the accruals based trading

strategy are not significantly different from zero, it would suggest that the accruals anomaly

does not exist in Sweden. By dividing the sample into two time periods, the study may

investigate if the anomaly has existed but been arbitraged away in later years. Furthermore, by

17

applying several methods of risk-adjustment, this study avoids being misleading with regards

to misspecified asset pricing models.

2.3.6 The accruals anomaly in Sweden

The research on the accruals anomaly on the Swedish market has reached ambiguous results.

When LaFond (2005) studied accruals’ implication on stock returns in 17 countries between

1989 and 2009, he found that the monthly returns of hedge portfolios where investors go long

in low accruals firms and short in high accruals firms generated significant abnormal returns

in Sweden. Using the calendar time monthly abnormal returns, he found that the hedge

portfolio had an alpha value of 0.69, significant at the 10% level. Even though the anomaly

was present in all but two countries studied, LaFond concluded that it is uncorrelated across

markets and therefore it is difficult to point out a common global source. (LaFond, 2005)

In contrast to LaFond’s significant results, Pincus et al. (2007), who studied 20 countries

between 1994 and 2002, did not find any support for the previous finding that Swedish

investors may earn abnormal returns by using an accruals based trading strategy. However,

there are some differences between the two studies worth pointing out. Firstly, LaFond (2005)

investigated a longer time period than the latter study. This results in a greater amount of

observations which may have improved the statistical tests. Secondly, when calculating

abnormal returns, LaFond (2005) used the more traditional Fama-French 3 factor approach

while Pincus et al. (2007) used a variation that accounts for earnings-to-price, size and book-

to-market. It is possible that the choice of model may have influenced the estimated abnormal

returns.

The authors of this paper have also found two master theses (Giaever & Gabrielsson, 2007;

Ivanov & Maximova, 2011) that investigate the returns from an accruals based trading

strategy on the Swedish stock market. The first thesis (Giaever & Gabrielsson, 2007) supports

the existence of the accruals anomaly in Sweden while the second one (Ivanov & Maximova,

2011) fails to find such evidence. However, it seems that the former study did not statistically

test their hedge portfolio returns. The ambiguous results on the Swedish market motivate for

introducing the new earnings benchmark in order to expand the knowledge in the area.

18

The relevance of previous research on the accruals anomaly in Sweden to this study

In case the results show that one can earn abnormal returns by trading on Swedish firms’

accruals, the findings would support LaFond (2005) and Giaever & Gabrielsson (2007).

Insignificant results would instead be in line with Pincus et al. (2007) and Ivanov &

Maximova (2011).

19

3. METHODOLOGY

The third chapter addresses the data collection process and methodology of the study. The

variables needed for the hedging portfolio will be defined as well as calculations of returns

and statistical tests. The chapter ends with a critical discussion of the methodology.

3.1 Data and sample selection

The primary source of raw data is Thomson Reuters Datastream. The study includes firms

listed on NASDAQ OMX Stockholm main market (Large cap, Mid cap and Small cap).

Financial firms such as banks, investment firms and insurance companies have, in line with

previous research, been excluded from the sample due to the special nature of their financial

statements. They are identified by using Thomson Reuters Datastream’s list of “Financial

Services” firms and double checked using the Global Industry Classification Standards

benchmark.

In order to reflect the information available to investors at the time of the portfolio formation,

portfolios are formed based on the prior year’s financial statement data. Portfolios are formed

in June as it is assumed that all firms will have completed their financial statements by then

and made them publicly available. For example, for the portfolio reaching from June 2002 to

June 2003, stock returns are recorded from June 2002 to June 2003 and the classification

variables are collected from the year-end financial statement data from 2001. Thus, firms

must have had financial statement data during any of the years between 2000 and 2011 and

historical stock price data for the subsequent year. However, firms need not have been listed

during the entire investigation period, but must have data available for at least one portfolio

formation. This selection is done to avoid the look-ahead bias determined in previous research

(Hafzalla et al., 2007; Kraft et al., 2006) and will be discussed in detail in section 3.1.1. By

forming the portfolios in this way, the trading strategy becomes realistically implementable,

adding to the practical applicability of the results.

As all data has been collected from Thomson Reuters Datastream, the study is vulnerable to

any faults in the database. However, as Datastream is widely used in research and in practice,

it may be assumed that the amount of incorrect data is likely low. Furthermore, all

calculations, apart from the Calendar time regressions, have been performed manually in this

study. Thus, there is room for human error in the data. However, the authors have been

20

careful to double-check all calculations and portfolio formations in order to minimize the

risks for mistakes.

3.1.1 Biases in accrual research

Kraft et al. (2006) identify a number of common biases that show up in several previous

studies on the accruals anomaly. One of the most common ones is the look-ahead bias. This

refers to the recurring requirement in research papers that sample firms must have reported

earnings in the year t+1 (i.e. the year after portfolio formation) and is present in studies

performed by Sloan (1996), Collins & Hribar (2000), Xie (2001) among others. The reason

for this sample selection is often that the researchers also investigate the persistence and

pricing of accruals, which requires that next year’s earnings are available. However, when

only investigating the trading strategy, the selection is difficult to motivate intuitively as it is

impossible for investors to predict which firms will be delisted during the year and which will

remain on the market. The issue with the criteria is that it creates an upward bias in the

measure of abnormal returns, as firms included in the research are only the firms that have

survived during the event year. (Kraft et al., 2006) In fact, the bias seems to distort results

significantly as Hafzalla (2007) and Kraft et al. (2006) no longer get abnormal returns that are

significantly different from zero for the low accrual firms, when the look-ahead bias is

removed.

In order to account for the look-ahead bias, the authors of this study have gathered

information on which stocks have had financial statement data available during the data

collection period, but have been delisted during the research period 2002-2013. The financial

information and stock prices have then been collected from Thomson Reuters Datastream.

The delisted firms have been included in a portfolio if they have had stock returns for at least

the first month from the portfolio formation. Once they are delisted, the proceeds are invested

in the market index and not included in next year’s hedge portfolio.

3.2 Variables

This section discusses and defines the variables chosen to classify firms when forming

portfolios for the accruals based trading strategy. While there are a number of ways to

calculate each variable, the definitions in this paper will be similar to those in Hafzalla et al.

(2007). This is done in order to increase comparability.

21

3.2.1 Accruals

There are two methods commonly used in previous literature when estimating accruals; the

balance sheet approach and the cash flow approach. The balance sheet approach estimates

accruals based on changes in balance sheet line items and is calculated as follows:

(1)

Where:

ΔCA = change in current assets year y

ΔCash = change in cash /cash equivalents year y

ΔCL = change in current liabilities year y

ΔSTD = change in debt included in current liabilities year y

ΔTP = change in income taxes payable year y

Dep = depreciation and amortization expense year y

While this method is found in several studies conducted on the accruals anomaly (Sloan 1996;

Zach, 2003; LaFond, 2005; Kothari et al., 2006; Kraft et al., 2006; Green et al, 2011), Hribar

& Collins (2002) identify a misspecification and bias when using the balance sheet approach

to estimate accruals. The balance sheet approach relies on an assumed relationship between

the capital accounts on the balance sheet, and the accrual element of revenues and expenses

on the income statement. However, Hribar & Collins (2002) argue that when firms engage in

acquisitions, divestitures, accounting reclassifications etc., the non-operating events are

incorporated into the balance sheet accounts, but do not ”flow through” to the income

statement. If accruals are measured by the balance sheet approach, the balance sheet account

changes due to such events would incorrectly be considered a change in the accruals

component of earnings, leading to errors in measurement when using this approach. (Hribar &

Collins, 2002)

The alternative method of estimating accruals is by means of the cash flow approach. This

approach begins by determining the operating cash flow portion of earnings from the

statement of cash flows. The operating cash flows are then deducted from earnings to give an

accruals estimate:

(2)

22

Where:

Earnings = the firms earnings (can be defined in different ways, discussed in section 3.2.2)

Operating Cash flows = cash flows from firms operations (discussed in section 3.2.3)

Previous research has varied slightly on which measure is applied. Sloan (1996) uses the

balance sheet approach, primarily motivated by the fact that the disclosure requirements of

SFAS no. 95 (which requires the statement of cash flows to report the information necessary

to compute the accrual portion of earnings from the reconciliation of net income with

operating cash flows) had only been in place for 4 of the 30 years he was examining.

Accordingly, using the cash flow approach for those years would create an inconsistency in

the calculation across firms. However, several researchers (see for example Zach, 2003;

LaFond, 2005; Kraft et al., 2006; Hafzalla et al. 2007) have employed the cash flow method

when estimating accruals. Based on Hribar & Collins (2002) findings of the weakness of the

balance sheet approach, and to ensure comparability with Hafzalla et al. (2007), this paper

applies the cash flow approach, using the formula (2) above.

3.2.2 Earnings

Previous studies have defined earnings in various ways, the most common are net income

before extraordinary items and net income (for example Kraft et al., 2006; Hafzalla et al.,

2007,). When Hribar & Collins (2002) tested different measures and definitions of accruals,

they found that by defining total accruals as net income before extraordinary items minus

operating cash flow, the researcher is provided with the most comprehensive measure of a

firm's accruals. The measure includes accruals related to deferred taxes, restructuring charges

and special items, but excludes accruals related to extraordinary items and discontinued

operations. Hribar & Collins (2002) argue that this measure is least likely to give incorrect

accruals estimations. This study will therefore use net income before extraordinary items as

the measure for earnings2.

3.2.3 Cash flows from operations

In line with previous research, the data for operating cash flows is taken directly from the

statement of cash flows from Thomson Reuters Datastream. The variable includes net cash

2 The data is collected from Thomas Reuters Datastream and is measured after taxes.

23

receipts and payments resulting from the operations of the firm, including funds from and for

working capital.

3.3 Classifying firms according to their accruals using different

benchmarks

Once the accruals have been determined as outlined above, portfolios of firms are created

based on the size of accruals in order to test an accruals based trading strategy. As mentioned

earlier, the majority of previous research scale accruals by some measure of firms size (often

total assets) when classifying firms into portfolios. Sloan (1996) motivates the use of firm size

with that it allows for comparison cross-sectionally and intertemporally. Hafzalla et al. (2007)

however, introduce the benchmark of earnings as a repair to the accruals based trading

strategy and argue that it significantly improves the strategy's abnormal return. The new

benchmark is motivated with the argument that it focuses on the composition of earnings and

the percent that is constituted by accruals and thereby, intuitively linked to the notion of

earnings fixation. (Hafzalla et al., 2007) In accordance with Hafzalla et al. (2007) this study

uses both the benchmark of size and earnings to compare the returns of the strategy when

applied to firms listed on NASDAQ OMX Stockholm main market. When applying the size

benchmark, accruals in year y are scaled by average total assets in year y and y-1:

(3)

When applying the earnings benchmark on the other hand, accruals in year y are scaled by the

absolute value of net income before extraordinary items in year y:

(4)

The absolute value is used when scaling by earnings to ensure that the sign of accruals is

maintained when classifying firms into portfolios. The effect is that firms that generally are

sorted into the lowest accrual portfolio have large positive cash flows, but accruals have

lowered earnings so they are close to zero. In the meantime, firms generally sorted into the

highest accrual portfolio generally have negative cash flows but accruals have increased their

24

earnings so they are close to zero. As earnings can be close to or equal to zero, using the

absolute value of earnings gives rise to the possibility of having extremely positive and

negative values of percent accruals. However, the value of percent accruals is merely used for

classification purposes, and is not relevant when measuring the returns once the portfolios

have been created. (Hafzalla et al., 2007)

The size accruals and percent accruals values are calculated for each of the 11 years being

investigated, for all firms that have the data required to be included in the portfolio.

Datastream may not lack data from their latest annual report regarding earnings, operating

cash flows or total assets, as this would hinder a proper classification. Once the size accruals

and percent accruals values have been calculated, firms are ranked and sorted into portfolios.

This sorting is done for each benchmark separately. While previous researchers (Sloan, 1996;

Kraft et al., 2006; Hafzalla et al., 2007) sort firms into deciles, this study uses quartiles

because the sample is smaller than in many other studies. This is done to ensure that the

sample being investigated in the trading strategy, i.e. the highest and lowest quartiles, is not

too small to ensure statistical reliability. As this study includes firms that are delisted during

the investigation period, as well as firms that have been listed after 2002, the number of firms

in the quartiles vary between the years3

. In total, the study includes 2363 firm-year

observations. There are between 50 to 60 firms in each quartile each year, resulting in

approximately 581-598 firms in each quartile in the entire investigation period. Within each

quartile, the stock returns of each firm are computed to estimate the portfolio return. The

computation of returns and abnormal returns are discussed below.

3.4 Returns

3.4.1 Stock returns

Stock prices for each firm are recorded monthly from the end of June in the year that the

portfolio is formed (six months after the fiscal year end from which the financial statement

data has been collected). The stock price is defined as the adjusted stock price in Thomson

Reuters Datastream as this accounts for dividends and stock splits, giving a more accurate

3 Newly listed firms are only included in portfolios from the year when they fulfill the requirements regarding financial statement data and

stock price information (see requirements explained in section 3.1)

25

reflection of the return earned by an investor than, for example, the unadjusted closing price.

The monthly stock returns for each firm are computed as the relative change in stock price:

(5)

Where

Rit = is the return of the security i in month t

Pit = is the price of the security i in month t

Pit-1 = is the price of the security i in month t-1

3.4.2 Abnormal stock returns

When evaluating the return from an investment, it is important to recognize the relationship

between risk and return, which indicates that a higher return may be associated with the

investor having accepted a higher risk. This makes it inappropriate to assess the raw return

and instead motivates the use of risk adjustment in order to estimate the abnormal return of an

investment.

It is common to use a historical estimation period to estimate the normal returns, and then

estimate abnormal returns as the difference between realised and historical normal returns.

However, Sloan (1996) argues that this method is inappropriate when firms are classified

according to characteristics that are unstable, such as accruals. This is because using a

historical estimation period implicitly assumes that the relative risk of portfolios remain the

same over time, which may be unlikely. By using a method that estimates normal returns at

the same time as the actual returns, the risk is permitted to change over time, which is

preferable. There are two commonly used ways of risk adjusting long-run stock returns in

previous accruals literature that accomplish this. The first method calculates the buy-and-hold

abnormal returns using carefully constructed and individually matched size reference

portfolios. The second estimates calendar time abnormal returns for sample firms. As

previous researchers have used different approaches, and in order to robustness check the

results, this study employs both.

When estimating and risk-adjusting long run abnormal stock returns, Barber & Lyon (1997)

argue that there are three common biases that may cause misspecified test statistics. These

are:

26

1. The new listing bias: refers to a misspecification as the firm return is compared to an

expected return proxy (for example a reference portfolio or market index), where the

firm's stock price reflects a long history of events but the expected return portfolio

includes new firms that begin trading subsequent to the event month.

2. The rebalancing bias: occurs when estimating the reference portfolio return. When the

returns of all firms in the market or reference portfolio are compounded to find the

return of the reference portfolio, it is often assumed that the weights of each firm are

rebalanced monthly. In the meantime, no such rebalancing is done to the sample firm,

causing a bias in the abnormal return estimation.

3. The skewness bias: refers to the empirical finding that long-run abnormal returns are

positively skewed.

Generally, the bias related to the rebalancing bias and skewness bias is negative, while the

bias related to new listings is positive (Barber et al., 1999) This should be taken into account

when estimating abnormal returns and will be discussed in relation to this study below.

3.4.3 Buy-and-hold abnormal returns using reference portfolios

Estimating the abnormal returns using the size-adjusted buy-and-hold return entails

classifying all firms by their market capitalization4 on the last day of the year ending before

the year when the portfolio is formed. For example, for the portfolio formed in 2002, firms

are ranked according to their market capitalization on the last of December in 2001. Based on

their market capitalization, firms are sorted into 10 portfolios. Firms with relatively high

market capitalization are sorted into portfolio 10, while firms with relatively low market

capitalization are sorted into portfolio 1. While there is compelling evidence that market-to-

book also systematically accounts for a portion of risk and return (Fama & French, 1993),

Barber et al. (1999) find little difference when also accounting for market-to-book when

assigning reference portfolios. As all portfolio creations, calculations and portfolio matching

is done manually in this study, only the market capitalization is considered when creating

reference portfolios.

4 Market Capitalization is defined as market price at year end multiplied by the number of shares outstanding and represents the total market

value of the firm. The variable is collected from Thomson Reuters Datastream.

27

The return calculation for each reference portfolio can be done by finding the average stock

return of all firms in the portfolio for each month, and then compounding the monthly average

return for the 12 month event window. However, when applying this calculation it is

implicitly assumed that the reference portfolio is rebalanced monthly, giving rise to the

rebalancing bias discussed above. It also leads to the new listing bias as firms listed after the

portfolio formation are included in the reference portfolio. Instead, Barber et al. (1999) argue

that the reference portfolio return should be calculated by compounding the monthly returns

of each firm in each size portfolio for the 12 month window individually, and then averaging

across all firms in the size portfolio:

(6)

Where:

is the buy-and-hold return of the size reference portfolio ƿ for time period τ

S is the first month of trading

τ is the total buy-and-hold period (in this case 12 months)

np is the no. of firms in the size portfolio ƿ,

R itƿ is the return of the security i , that belongs to size portfolio ƿ, in month t

By applying formula (6), Barber et al. (1999) argue that the return on the reference portfolio

represents “a passive equally weighted investment in all securities constituting the reference

portfolio” (Barber et al., 1999, p. 169) during the event window. As each security is given

equal weights once the compounded returns are calculated, there is no monthly rebalancing in

this method, effectively eliminating the rebalancing bias. Furthermore, as the investment in

the securities is done at the beginning of the portfolio formation, firms listed after the

formation are not included, effectively eliminating the new listing bias. The skewness bias is

addressed in the test statistics by applying a skewness adjusted t-test when testing abnormal

returns.

Once the total buy-and-hold return for each size reference portfolio has been estimated using

the formula above, the abnormal returns for each security are estimated as follows:

(7)

28

Where:

BHARi τ ƿ is the buy-and-hold abnormal return for firm i, belonging to size portfolio ƿ, for the period τ

R i t ƿ is the return of the security i, that belongs to size portfolio ƿ, in month t

is the buy-and-hold return of the size reference portfolio ƿ for time period τ

Once the buy-and-hold abnormal returns for each security have been calculated using formula

(7) the return of each accruals portfolio is estimated using the formula below. This is done for

each quartile in both benchmarks.

(8)

Where:

BHARqτ is the buy-and-hold return of quartile q for the period τ

BHARiτ is the buy-and-hold abnormal return for firm i for the period τ

nqτ is the no of firms in quartile q for the period τ

The hedge portfolio is constructed by taking a long position in the low accruals quartile and a

short position in the high accruals quartile. Thus, the return from the hedge portfolio is

calculated as follows:

(9)

Where:

BHARHs is the buy-and-hold return of the hedge portfolio that begins in month s

BHARSs is the buy-and-hold return of the short position (quartile 4) taken in month s

BHARLs is the buy-and-hold return of the long position (quartile 1) taken in month s

Finally, the significance of the buy-and-hold abnormal return in each quartile as well as in the

hedge portfolio is tested. While this study employs the conventional test statistic in order to

ensure comparability with Sloan's (1996) results from the buy-and-hold abnormal returns, the

skewness adjusted t-statistic introduced by Johnson (1978) is also applied. The reason for

applying a skewness adjusted t-statistic is that Barber & Lyon (1997) identify a positive

skewness in long-term buy-and-hold abnormal stock returns, which is reflected in a

negatively biased t-statistic. This leads to the risk of p-values being lower than they should in

29

lower tailed tests, and p-values being higher than they should be in upper-tailed tests (Barber

et al., 1999). The skewness adjusted test statistic is calculated as follows5:

(10)

Where:

and

tsa is the skewness adjusted t-statistic

is the mean buy-and-hold abnormal return for the sample

is the cross sectional standard deviation

n is the no. of observations

As in the conventional test statistic, the null hypothesis is that the mean abnormal returns are

not significantly different from zero. If the null hypothesis is rejected, the buy-and-hold

abnormal returns are significantly different from zero. Calculations for the buy-and-hold test

statistics are performed manually and summarized in Appendix 1: figure 1.

While Barber et al. (1999) make a convincing case for the use of the buy-and-hold abnormal

return using size matched reference portfolios, there is still a wide amount of researchers who

prefer the calendar time approach.

3.4.4 Calendar time approach

Fama (1998) and Mitchell & Stafford (2000), strongly advise using the calendar time

approach when estimating abnormal returns. Mitchell & Stafford (2000) argue that the

calendar time approach's strength lies in that it is robust to most statistical problems and find

evidence against Loughran & Ritter's (2000) argument that the calendar time approach should

suffer from low power in assessing the reliability of abnormal returns. This is in line with

Barber et al. (1999) who argue that from a researcher's point of view, the calendar time

approach offers the advantage of being able to assess if the abnormal returns are persistent

and yield more robust test statistic in non-random samples than does the buy-and-hold

approach6. Because the sample firms investigated in this study are characterized by extreme

5 Barber et al. (1999) argue that the skewness adjusted t-statistic should also be bootstrapped; however, this is beyond the scope of this study.

6 The test statistics may still be misspecified, but less so than in the buy-and-hold case. (Barber et al., 1999)

30

accruals, the sample is by design non-random. This creates a strong argument for the use of

the calendar time approach.

The disadvantage of the calendar time approach is that it does not precisely mimic the return

earned by the investor during the entire one year event window, and is sensitive to the model

of asset pricing. (Barber et al., 1999) Barber et al. (1999) state that when testing if the

abnormal returns are equal to zero, the researcher is actually doing a joint test that the

abnormal returns are equal to zero, as well as, testing that the chosen asset pricing model is

correct and valid. To account for this, this study employs two asset pricing models, both the

Fama-French 3 factor model developed by Fama & French (1992), and the Sharpe-Lintner

version of the Capital Asset Pricing Model (CAPM) (Sharpe, 1964; Lintner, 1965) when

performing the calendar time regressions.

In this method, the portfolio return is defined as the average monthly returns of the stocks that

constitute each portfolio. Thus, there will be one monthly portfolio return for each quartile in

both benchmarks. Below is the formula for calculating the monthly portfolio return:

(11)

Where:

Rqτ is the return of the accruals quartile q in month t

Rit is the return of the individual security i in month t

nqτ is the number of firms in the accruals quartile q in month t

The return of the hedging strategy is the sum of the returns generated by taking a long

position in firms in the lowest accruals quartile and a short position in firms in the highest

accruals quartile:

(12)

Where:

RHt is the return of the hedge portfolio in month t

RSt is the return of the short position (quartile 4) in month t

RLt is the return of the long position (quartile 1) in month t

31

To estimate the monthly abnormal returns in this method, a calendar time regression is run to

generate an alpha value. The alpha is the estimate of the monthly calendar time abnormal

return. The calendar time regression based on CAPM is specified as follows:

(13)

Where:

α is the abnormal return of portfolio

Rpt is the return of the portfolio in month t (done for each individual quartile and entire hedge portfolio)

Rft is the risk-free rate estimated as the Swedish Treasury bill 30-day rate in month t

Rmt is the return on the market in month t7

The calendar time regression based on Fama-French three factor model is specified as:

(14)

Where:

α is the abnormal return of portfolio

Rpt is the return of the portfolio in month t (done for each individual quartile and entire hedge portfolio

Rft is the 1 month risk free rate estimated as the Swedish Treasury bill 30-day rate in month t

Rmt is the return on the market in month t

SMB is the return on a portfolio that assumes a long position in small market capitalization firms and a

short position in firms with a larger market capitalization.

HML is a portfolio where firms are divided by the market-to-book value. The portfolio return is the

return from the high MtB firms minus the return from the low MtB8

The monthly abnormal return using this method is the alpha value. Its significance is tested

using a conventional t-test. However, as the estimation is specified as an Ordinary Least

7 This study uses the return on the market index of Stockholm OMX collected from Thomson Reuters Datastream. The variable is chosen as

it is comprised of all firms on NASDAQ OMX Stockholm, and can therefore be assumed to be representative of the trends on the general

Swedish market.

8 When performing studies on the US market, researchers are provided with Fama-French factors from Kenneth French’s website. However, comparing returns to HML and SMB portfolios constructed using US firms is inappropriate when investigating Swedish firms. As estimating

and constructing Fama-French factor portfolios can be time-consuming, this study has used values for the HML and SMB variable that are

provided by Professsor Stefano Marmi on http://homepage.sns.it/marmi/Data_Library.html. The methodology for constructing portfolios is identical to the one in Fama & French (1993).

32

Squares (OLS) regression, the OLS assumptions are tested for each regression in order to

ensure that correct inferences are drawn. The assumptions are:

1. The dependent variable is a linear function of the explanatory variables. This is

essential in order to detect if the functional form is correct. It is tested using the

RESET test in Eviews, where the rejection of the null hypothesis implies that the

model is misspecified and/or contains non-linearity. (Brooks, 2008)

2. The average value of the error terms is zero. This is tested in Eviews by

running the RESET test, however as there is a constant included in the regression, this

condition is assumed to hold. (Brooks, 2008)

3. : The variance of the error terms is constant and finite. Even

though the presence of heteroskedasticity may yield consistent and unbiased results,

the regression model would be inefficient and the standard errors may be misleading,

leading to incorrect inference. The assumption is tested by running White’s

heteroskedasticity test. If the null hypothesis is rejected, there is heteroskedasticity

which may be corrected for using White's robust standard errors in Eviews. (Brooks,

2008)

4. : Residuals are uncorrelated with another over time. In order to

investigate the assumption of positive autocorrelation the Durbin-Watson test is

performed. If the null hypothesis is rejected there is positive autocorrelation between

the residuals, which could cause the same consequences as the presence of

heteroskedasity, namely incorrect inference. As this study investigates stock returns

over time, the probability of autocorrelation is high. (Brooks, 2008)

5. : The independent variables are non-stochastic and not highly

correlated with each other. However, as long as these independent variables are not

correlated with the residuals, evidence shows that OLS will still be consistent and

unbiased even if the x independent variables are stochastic. The assumption of no

multicollinearity is tested using a correlation matrix, where the correlation between

variables should not be above 0,8. (Brooks, 2008)

6. : The error terms are normally distributed. This is investigated by

running a Jarque-Bera test of normality as well as plotting histograms. (Brooks, 2008)

The corrections that are made to each regression are presented in Appendix 2: figure 2 & 3.

33

3.4.5 Winsorization

As Kothari et al. (2006) finds that the previous findings of the accruals anomaly are largely

attributable to outliers, this study will investigate if the results are sensitive to extreme

observations. Winsorization is a process that attempts to deal with extreme outliers in a

research sample. It entails setting the values of extreme observations equal to the value of a

chosen percentile cut off. For example, when winsorizing by 5%, the observations below the

5th

and above the 95th

percentile are assigned the value of the 5th

and 95th

percentile

respectively. (Kothari et al., 2005) The effect is that outliers are brought closer to other

observations that are assumed to represent the true distribution of a population. The argument

for using such a data transforming method is to avoid the research results being incorrectly

influenced by spurious outliers. It also offers a robustness check to see if results are

attributable to extreme observations. (Kothari et al., 2005) Winsorization can be performed

for various percentile cut offs, such as 0.1%, 1% or 5%. The disadvantage of using such a

method is that it transforms the data to no longer reflect the actual observations, making the

findings less generalizable. Kothari et al. (2005) argue that trimming (i.e. completely

excluding extreme observations) or winsorizing may lead to biased inferences due to the non-

random deletion from the sample.

While the trading strategy becomes less realistically implementable when winsorizing the

returns, the findings of Kraft et al. (2006) motivate need for testing the results once outliers

have been accounted for. This is considered a complement to the regressions on the raw data.

The reason for choosing to winsorize the sample instead of trimming it, is because trimming

completely eliminates the extreme values, which may be more misleading than when

transforming them. The results of the abnormal returns using the winsorized data are

presented in the next chapter.

34

4. RESULTS AND ANALYSIS

This chapter begins by comparing the characteristics of the firms when applying the total

assets and earnings benchmarks. The results of the accruals based trading strategy are then

presented and discussed in relation to previous research.

4.1 Comparing the benchmarks

To investigate if the benchmarks differ when classifying firms into extreme accrual portfolios,

firm characteristics in each quartile are compared within each benchmark as well as between

the benchmarks. Appendix 3 presents the average and median values for market

capitalization, stock price, cash flows from operations, net income and total assets for the

firms in each benchmark’s quartiles. As the focus is to identify common characteristics of the

majority of the firms within each quartile, the analysis will focus on the median value as the

mean value may be influenced by a few extreme firms. Upon examination, it appears that the

benchmarks often classify the same firms into quartile 4. Each year, only a few firms that are

classified into quartile 4 using the total assets benchmark are not classified into the same

quartile using the earnings benchmark. The firms sorted into the 1st quartile differ to a greater

extent and are therefore in greater focus in this comparison.

4.1.2 Market capitalization

The median quartile market capitalization is illustrated in figure 4a below:

Figure 4a

When studying the quartiles within the total asset benchmark, quartile 1 is most noteworthy as

the median market capitalization of these firms is substantially lower than in the other

quartiles. Instead, the largest firms seem to have been classified into quartile 2 and 3. As size

is negatively correlated with risk, and risk is positively correlated with returns (Fama &

French, 1992), the potential returns in quartile 1 using the total assets benchmark (that

35

contained smaller firms) can be assumed to be higher than in the other quartiles. On the other

hand, a higher portion of cash flows in earnings seem less risky as these are more persistent,

causing earnings to be more persistent. When assessing the quartiles in the earnings

benchmark, the market capitalization is more even across quartiles.

When comparing the two benchmarks it is obvious that firms in quartile 1 are larger when

sorting by earnings than sorting by total assets. The median market capitalization in quartile 1

using the earnings benchmark is 864mSEK while the median using the total assets benchmark

is 459mSEK. Thus, firms sorted into quartile 1 by the earnings benchmark are almost twice as

big as firms sorted into quartile 1 by the total assets benchmark. The results are in line with

Hafzalla et al. (2007). However they are not as extreme as Hafzalla et al. (2007) found that

firms in quartile 1 sorted by the earnings benchmark are four times larger than the firms

sorted by the total assets benchmark.

4.1.3 Stock price

The median quartile stock price is illustrated in figure 5a below:

Figure 5a

As may be intuitive after investigating market capitalization, firms in quartile 1 have the

lowest stock price (26.74 SEK) when sorting by total assets benchmark. The highest stock

price is found in quartile 3 (44.4 SEK.) The stock price in quartile 2 is 42.4 SEK and in

quartile 4 is 33.69 SEK. As Kraft et al. (2006) state that relatively low stock prices suggest

poor past performance, it seems counterproductive that the firms with the lowest stock prices

are assigned to quartile 1. This is because the trading strategy consists of taking a long

position in quartile 1 and a short position in quartile 4, making it preferable for bad

performing firms to end up in the high accrual quartile where the investor profits from a

36

declining stock price. Using the earnings benchmark on the other hand, the median stock price

is lowest in quartile 4 (32.94 SEK). As mentioned, poor performance is desirable when

shorting firms and thus the earnings benchmark seems to sort firms into quartile 4 effectively

with respect to their stock price. Meanwhile, the median stock price in quartile 1 is 34.17 SEK

and 39.85 SEK in quartile 2. Again, firms in quartile 3 have the highest stock price: 43.67

SEK. The fact the middle quartiles contain firms with the highest median stock prices using

both benchmarks suggests that the majority of well performing firms end up in these quartiles.

In combination with significant positive abnormal returns found in table 1a and 2a using the

Fama-French 3 Factor model, it suggests that it may be more likely to earn abnormal returns

by taking a long position in these firms instead of focusing on firms with extreme accruals.

The relative transaction costs decrease with higher stock prices (Hafzalla et al. 2007). Thus

the cost of taking a long position in low accrual firms would seem to be lower when firms are

sorted by earnings rather than total assets. As the stock prices in quartile 4 are relatively

similar, there is likely little difference in transaction costs arising from taking a short position

in high accrual firms. All in all, with respect to transactions costs, the earnings benchmark

appears to be superior to the total assets benchmark.

4.1.4 Net income and Cash flows from operations

The median quartile net income and cash flows from operations are illustrated in figure 6a &

7a below:

Figure 6a

37

Figure 7a

The median net income as well as cash flows from operations are presented in figure 6a and

7a respectively. The quartile characteristics of net income and cash flow from operations are

in line with Hafzalla et al. (2007). For the total assets benchmark the median net income in

quartile 1 is -9mSEK while the median cash flows are 48mSEK. For quartile 4, the opposite is

found as cash flows are low but net income is high. For the earnings benchmark, the median

net income is 11mSEK, and the median cash flows from operations amount to 106mSEK.

Again, the opposite is found for firms in quartile 4. Intuitively, this is expected as high and

low accruals respectively, are the primary characteristics used to classify firms into portfolios.

The median net income is relatively low in the lowest accrual portfolio for both

benchmarks. The median cash flows are substantially higher in quartile 1 using the earnings

benchmark, suggesting that the earnings benchmark is better at sorting firms with both higher

net income and, more importantly, cash flows, into quartile 1. From an investor perspective

this may be important as larger cash flow components are assumed to be associated with a

higher earnings quality (Bernstein, 1993). Thus, the high cash flows coupled with the low

stock prices in this quartile that were discussed above, may be an indication that the firms in

quartile 1 are undervalued.

4.2. Results from testing the existence of the accruals anomaly in Sweden

In order to investigate if the accruals based trading strategy generates abnormal returns on the

Swedish stock market, the abnormal returns for portfolios based on the total assets and

earnings benchmark are estimated. As discussed in chapter 3, when estimating the abnormal

returns, this study employs the calendar time approach using the CAPM and Fama-French 3

factor model as well as the size-adjusted buy-and-hold approach.

38

4.2.1 Total assets benchmark

To begin with, the total assets benchmark is investigated. Table 1a below shows the results

using the calendar time approaches.

Table 1a. Calendar Time Approach to Total Assets Benchmark

Abnormal returns using the calendar time approach for the total assets benchmark. The top value represents the

time series means of abnormal return (alpha) and the number below is the p-value corresponding to the t-test.

Quartile Return CAPM Return Fama-French 3 factor

Quartile 1 0.003 0.005

0.46 0.08*

Quartile 2 0.002 0.004

0.48 0.07*

Quartile 3 0.001 0.003

0.58 0.12

Quartile 4 0.002 0.004

0.41 0.06*

Hedge portfolio (Q1- Q4) -0.007 -0.006

0.21 0.29

* significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level

As can be seen from Table 1a, the monthly abnormal returns from the hedge portfolio are not

significantly different from zero for either of the risk adjustment measures. Thus, the results

do not support the findings of LaFond (2005) and Giaever & Gabrielsson (2007) that an

accruals based trading strategy may lead to abnormal returns in Sweden. The findings instead

seem to support the results found by Ivanov & Maximova (2011) and Pincus et al. (2007).

When looking at the individual quartiles in table 1a, the monthly abnormal returns estimated

by CAPM yield no significant results. On the other hand, when applying the Fama-French 3

factor model to the calendar time approach, quartiles 1, 2 and 4 yield positive abnormal

returns at the 10% level. However, the 10% significance level should be considered with

caution and is not preferable when assessing the statistical reliability (Brooks, 2008). To

further investigate the results of the total assets benchmark, the size-adjusted buy-and-hold

abnormal returns are presented in table 1b below:

39

Table 1b. Buy-and-Hold Approach to Total Assets Benchmark

Abnormal returns using Buy-and-Hold size matched reference portfolios for the total assets benchmark.

Quartile Mean return Conventional t-statistic Skewness adjusted t-statistic

Quartile 1 -0.006 -0.413 -0.395

Quartile 2 0.000 -0.058 -0.038

Quartile 3 0.004 0.238 0.247

Quartile 4 0.001 0.039 -0.014

Hedge portfolio (Q1-Q4) -0.007 -0.246 -0.200

The critical t-value using the standard t-distribution is 2.228 at the 5% significance level and 1,812 at the 10% level.

Table 1b above shows the mean buy-and hold returns as well as the t-statistics for the

conventional t-test and skewness adjusted t-test. When comparing the t-statistics it is clear

that the skewness adjusted t-statistic is slightly higher. This suggests that the skewness

adjusted t-statistic has corrected for the positive skewness in long-term buy-and-hold

abnormal stock returns, that results in a negatively biased t-statistic (see chapter 3). However,

neither t-statistic is higher than the critical t-value, indicating that the mean buy-and-hold

abnormal returns cannot be concluded to be significantly different from zero.

All in all, the results from all three risk adjustment measures indicate that implementing an

accruals based trading strategy using the total assets benchmark, does not yield consistent

abnormal returns in Sweden.

4.2.2 Earnings benchmark

The investigation continues by evaluating the earnings benchmark proposed by Hafzalla et al.

(2007). The results from the calendar time approach using the CAPM and Fama-French 3

factor model are presented in Table 2a.

40

Table 2a. Calendar Time Approach to Earnings Benchmark

Abnormal returns using the calendar time approach for the earnings benchmark. The top value represents the

time series means of abnormal return (alpha) and the number below is the p-value corresponding to the t-test.

Quartile Return CAPM Return Fama-French 3 factor Quartile 1 0.002 0.004

0.43 0.10*

Quartile 2 0.002 0.004

0.46 0.05**

Quartile 3 0.002 0.004

0.41 0.05**

Quartile 4 0.002 0.004

0.54 0.09*

Hedge portfolio (Q1- Q4) -0.007 -0.007 0.22 0.23

* significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level

In similarity with the total assets benchmark, the returns of the hedge portfolio are

insignificant for both risk adjustment measures. This suggests that Hafzalla et al.’s (2007)

benchmark has not improved the returns from the hedge portfolio. When investigating the

quartiles individually, the returns are once again insignificant when applying the CAPM. The

statistical significance when using Fama-French 3 factor model has however improved. The

returns are then significantly positive at the 5% level for quartile 2 and 3 and at the 10% level

for quartiles 1 and 4. Overall, the results from The Fama-French estimation suggest that there

is little difference in abnormal returns between the quartiles, as they on average appear to be

positive. To see if the size-adjusted buy-and-hold abnormal returns yield similar results, the

findings are presented in Table 2b below:

41

Table 2b. Buy-and-Hold Approach to Earnings Benchmark

Abnormal returns using Buy-and-Hold size matched reference portfolios for the earnings benchmark.

Quartile Mean return Conventional t-statistic Skewness adjusted t-statistic

Quartile 1 0.007 0.272 0.355

Quartile 2 0.006 0.274 0.297

Quartile 3 -0.007 -0.335 -0.422

Quartile 4 -0.007 -0.460 -0.551

Hedge portfolio (Q1-Q4) 0.015 0.469 0.511

The critical t-value using the standard t-distribution is 2.228 at the 5% significance level and 1,812 at the 10% level.

Table 2b above presents the mean buy-and hold returns as well as the t-statistics for the

conventional t-test and the skewness adjusted t-test. When comparing the t-statistics, the

skewness adjusted t-statistic is slightly higher for the hedge portfolio as well as quartile 1 and

2. However, in similarity to the total assets benchmark, the mean buy-and-hold abnormal

returns cannot be concluded to be significantly different from zero for either the quartiles nor

the hedge portfolio. Again, the results from all three risk adjustment measures indicate that

the accruals based trading strategy does not yield significant abnormal returns on the Swedish

stock market.

4.2.3 Comparing the results to previous research9

The results in relation to previous studies supporting the accruals anomaly

As presented in the beginning of this paper, several researchers have found that the accruals

based trading strategy yields positive abnormal returns to an investor (Sloan, 1996; Collins &

Hribar, 2000; Bradshaw et al., 2001; Xie, 2001; Zach, 2003; LaFond, 2005; Hafzalla et al.,

9 As the results of both benchmarks yield similar returns they are analysed together in relation to previous research.

42

2007; Pincus et al., 2007). Sloan (1996) finds that the mean size-adjusted buy-and-hold

abnormal returns from the accruals based trading strategy is 10.4%. He attributes this to the

earnings fixation hypothesis and argues that investors fail to identify the components of

earnings which cause them to incorrectly price firms’ future earnings potentials. When

looking at the results for the individual quartiles in this study, the average monthly abnormal

returns for quartile 1 (table 1a and 2a) appear to be positive when applying the calendar time

approach with Fama-French 3 factor model. This suggests that the findings in quartile 1 are in

line with the Sloan’s (1996) earnings fixation hypothesis as it predicts that investors in low

accrual firms are positively surprised by the persistence in earnings, resulting in positive

abnormal stock returns. On the other hand, the positive abnormal returns for the 4th quartile

do not support this theory. This is because high accrual firms are expected to experience

negative abnormal returns as a result of investors being unable to acknowledge that accruals

are less persistent in predicting earnings than cash flows, and are therefore surprised when the

future earnings are lower than expected. Similarly, the results for the hedge portfolio return

not either support Sloan’s findings as the abnormal returns of the hedge portfolio cannot be

said to be significantly different from zero using any of the three risk-adjustment measures.

A possible explanation for the similar positive abnormal returns in quartile 1 and 4 and thus

divergence from the earnings fixation hypothesis, could be that the firms in quartile 4 appear

to have similar earnings characteristics as those in quartile 1. Based on the classifications,

firms in the high accrual portfolio are expected to have relatively low cash flow component in

earnings and a relatively high accruals component, while those in quartile 1 are expected to

experience the opposite. However, as discussed in section 4.1, it is clear that quartile 1 and 4

have the expected characteristics (compare Appendix 3: figure 6a &7a). Thus the similarity in

returns in quartiles 1 and 4 cannot be explained by the quartile firm characteristics. An

alternative explanation for the positive return in quartile 4 is that the level of accruals do not

reverse to the same extent in Sweden as in the US market, and instead appear to continuously

remain at high levels. If investors then continue to focus on earnings, and price the firms

accordingly, then maintaining high accruals will likely continue to ensure positive returns on

the stock. A third explanation may be that investors do recognize the difference in accruals,

and initially overreact to the high accruals in quartile 4’s firms causing negative returns.

However, they are later positively surprised when they realize that they have reacted too

strongly, leading to subsequent positive abnormal returns. An initial negative overreaction,

43

followed by positive returns in the following months, may explain the positive monthly mean.

It can also explain the low significance of the statistical tests as the tests are done on all

months in a year, and do not distinguish between the returns during the beginning and the end

of the year. Finally, one should keep in mind that the results for the individual quartiles

discussed above are only found using one of the three risk-adjustment methods. Thus, the

results do not hold for the size-adjusted buy-and-hold abnormal returns or the calendar time

approach with the CAPM and are therefore not robust to different methods of risk-adjustment.

An alternative theory to the earnings fixation hypothesis is the agency theory of overvalued

equity. Kothari et al. (2006) found that the return from the high accrual firms had -7.6% in

abnormal returns but that low accruals firms’ abnormal returns were insignificant. They

argue that the agency theory of overvalued equity was a possible explanation to the finding as

managers of overvalued firms had an incentive to adjust earnings upward with the use of

accruals, but that managers of undervalued firms did not have the same incentive to use

accruals to adjust earnings downward. The abnormal returns in high accrual portfolio are

therefore predicted to be negative and significant, while the returns in the low accrual

portfolio are expected to be insignificant. As the results in this study show significant positive

returns in quartile 1 when using the Fama-French factor adjusted calendar time approach

(table 1a and 2a) and the positive abnormal returns for quartile 4, this study does not support

the agency theory of overvalued equity explanation. However, if the agency theory of

overvalued equity would be accepted as the reasoning behind the accruals anomaly, the

results in this study suggest that the overvalued firms with high accruals continue to be

overvalued and generate abnormal returns to the investor. If this is the case, then the low

persistence of accruals should not be considered an issue as positive returns can be maintained

in spite of high accruals. However, as the results are not robust to different measures of risk-

adjustment, such a conclusion cannot be drawn.

Hafzalla et al. (2007) find that when sorting firms by the total assets benchmark, the low

accrual portfolio as well as the hedge portfolio is insignificant. The high accrual portfolio

return is significantly negative, -6.3%. However, when instead sorting by earnings, all returns

are significant at the 1% level. In their study, the low accruals portfolio yields an abnormal

return of 5.4%, the high accruals portfolio, -6.9% and the hedge portfolio, 12.4%. The results

in this study are not in line with Hafzalla et al. (2007) as the abnormal returns for the quartiles

44

have varying significance and sign, and the earnings benchmark does not seem to improve the

abnormal returns that may be earned from the accruals based trading strategy. As discussed in

section 4.1.2 of the analysis, Hafzalla et al. (2007) found that firms in quartile 1 sorted by the

earnings benchmark are four times larger than the firms sorted by the total assets benchmark.

This difference is substantially larger than the classifications in this study. While the relative

size of the firms may impact the potential and actual returns in the portfolios, it does not

explain the difference between the results in this study and Hafzalla et al. (2007). This is

because the methods of risk-adjustment, buy-and-hold abnormal return using size matched

reference portfolios (used in both studies) and the calendar time approach using the Fama-

French 3 factor model (used in this study) control for the size-effect on returns.

LaFond (2005) finds that the monthly abnormal return from the hedge portfolio in Sweden

has an alpha value of 0.69, significant at the 10% level. LaFond (2005) applies the calendar

time approach using the Fama-French 3 factor model, as is done in this study. The difference

in results may be due to the fact that LaFond (2005) investigates a longer time period.

However, in spite of the longer time frame, LaFond’s study only attains a low significance

level and the results in his study should be considered with caution. Together, the low

significance in LaFond (2005) and the lack of significance in this study suggest that the

accruals based trading strategy does not yield persistent monthly abnormal returns in Sweden.

Furthermore, this study does not support Giaever & Gabrielsson (2007) who found that it was

possible to earn abnormal returns by forming a hedge portfolio based on accruals. However,

Giaever & Gabrielsson (2007) use a different methodology compared to this study and

estimate abnormal returns based on historical values of the CAPM-model which may explain

the difference in the results. Additionally, it seems that Giaever & Gabrielsson (2007) only

calculate the mean abnormal returns to the hedge portfolio but fail to perform statistical

significance tests on the data. This makes it difficult to compare the results and draw any

further conclusions.

Khan (2008) and Wu et al. (2009) argued that the accruals anomaly could be largely

attributed to misspecified risk adjustments. In this study, the abnormal returns of the hedge

portfolio are insignificant when applying the calendar time approach using CAPM and Fama-

French 3 factor model, as well as when applying the size-adjusted buy-and-hold abnormal

returns. However, by applying Khan (2008) and Wu et al.’s (2009) reasoning, it may be

45

expected that at least one of the risk adjustment measures employed in this study would show

support of the existence of the accruals anomaly. A possible explanation for the failure to find

evidence in support of the anomaly would be if the mispricing was too small and short lived

for any of the risk adjustment measures to capture it. However in such a case, it may be

discussed if the anomaly is present, as the market corrects for it relatively quickly.

The results in relation to previous studies contradicting the accruals anomaly

The efficient market hypothesis (Fama, 1970) suggests that financial markets today are semi-

strong were all public information, such as accounting data, are incorporated into firms’ stock

prices. As discussed in chapter 2, evidence against the hypothesis is market anomalies, states

of inefficiency and mispricing. However, as this study does not find significant abnormal

returns for the hedge portfolio with any of the risk adjustment methods, there is no evidence

of market inefficiency due to mispricing of accruals. Thus, with respect to firms’ accruals and

earnings components, the Swedish stock market does not seem to be inefficient.

Pincus et al. (2007) finds that the accruals based trading strategy does not yield significant

abnormal returns in Sweden. They partly attribute this to their empirical finding that the

accruals anomaly is less likely to be present in civil law countries. However they fail to

theoretically motivate this finding as common law countries are generally found to be more

efficient, leaving less room for anomalies to appear (Hung, 2001). Still, as Sweden has a civil

law tradition, the results of this study are in line with the findings in Pincus et al. (2007).

Pincus et al. (2007) also suggest that the accruals anomaly is more likely to occur in countries

where investors pay little attention to the components of earnings. As Swedish firms on

NASDAQ OMX Stockholm main market are required to publish cash flow statements, the

information needed to determine the accruals component is made easily accessible. The high

transparency may increase the attention awarded to the components of earnings. By this

reasoning investors are more likely to price firms accurately with respect to their accruals,

making it is less likely for the anomaly to appear. The easy access to information in Sweden

may explain the lack of abnormal returns in this study. A final finding by Pincus et al. (2007)

is that the accruals anomaly is more likely to be present in countries with strict laws against

insider-trading. The reason that strict insider trading laws would give rise to the accruals

anomaly is that insiders are likely to have more extensive information on the future earnings

possibilities of the firm, which outsiders lack. When insiders are prohibited from trading on

46

their non-public information, it would suggest that mispricing is not arbitraged away as

quickly as the market lacks the insider’s extensive information. However, Pincus et al. (2007)

find that Sweden is considered to have relatively strict regulations on insider trading. As this

study finds that the accruals anomaly is not present in Sweden in spite of this, the results do

not to support Pincus et al. (2007) in this regard. Instead, it seems that the investors’ access to

information outweighs the information gap between insiders and the market.

The results of this study are in line with Green et al. (2011) who find that the accruals based

trading strategy does not yield abnormal returns. In particular, the findings by Green et al.

(2011) indicate that the abnormal returns generated by an accruals based trading strategy on

the US market seem to have declined after 2004. They attribute this to the extensive research

on the subject and the widespread implementation by practitioners, which have caused the

accruals anomaly to be arbitraged away. To test if the results of this study support the

predictions made by Green et al. (2011), the sample is divided into two sub periods; one

ranging from June 2002 to June 2008 and one ranging from June 2008 to June 2013. The tests

are only performed using the calendar time approach as the buy-and-hold abnormal returns

only generate one yearly mean, which makes the sample too small to ensure statistical

reliability when the sample is divided. The results from the calendar time approach using

CAPM and Fama-French 3 factor model are shown in the table 3a and 3b below:

Table 3a. Total Assets Benchmark (Sub periods)

Abnormal returns for sub periods using the calendar time approach for the total assets benchmark. The top value represents the time series

means of abnormal return (alpha) and the number below is the p-value corresponding to the t-test.

Quartile CAPM Fama-French 3 factor CAPM Fama-French 3 factor

2002-2008 2008-2013

Quartile 1 0.008 0.008 -0.003 0.001

0.18 0.05** 0.47 0.67

Quartile 4 0.004 0.004 0.000 0.003

0.33 0.21 0.99 0.36

Hedge portfolio (Q1-Q4) -0.022 -0.021 -0.01 -0.012

0.00*** 0.00*** 0.26 0.13

* significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level

47

Table 3b. Earnings Benchmark (Sub periods)

Abnormal returns for sub periods using the calendar time approach for the earnings benchmark. The top value represents the time series

means ofabnormal return (alpha) and the number below is the p-value corresponding to the t-test.

Quartile CAPM Fama-French 3 factor CAPM Fama-French 3 factor

2002-2008 2008-2013

Quartile 1 0.005 0.004 -0.002 0.002

0.18 0.09* 0.63 0.57

Quartile 4 0.005 0.005 -0.001 0.002

0.26 0.14 0,75 0.55

Hedge portfolio (Q1-Q4) -0.027 -0.026 -0.008 -0.010

0.00*** 0.00*** 0.37 0.19

* significant at the 10% level, ** significant at the 5% level, *** significant at the 1 % level

As shown in table 3a and 3b above, the return for the hedge portfolio during the time period

2002-2008 is significant when using the Fama-French 3 factor model, however the returns are

negative. For the period 2008-2013 the hedge portfolio returns are insignificant. As the results

are significant for the earlier period and insignificant for the later period, the results initially

seem to support that the accruals anomaly has been arbitraged away. However, as the hedge

portfolio returns for the period 2002-2008 have a negative alpha value of roughly -2% for

both benchmarks and both the CAPM and Fama-French risk adjustment, the results do not

support Green et al. (2011). Furthermore, in the CAPM estimation, it seems puzzling that the

returns for the hedge portfolio are negative while the returns for the individual quartiles are

not significantly different from zero. However, looking at it mathematically, when regressing

the hedge portfolio returns on the market excess return, the risk-free rate is deducted from the

raw hedge portfolio return (i.e. the return from quartile 1 minus the return from quartile 4).

Thus, if the returns from quartiles 1 and 4 cannot be said to be significantly different from

zero, the risk free rate will be deducted from zero, giving a negative hedge portfolio return. In

the case of the Fama-French estimation, the returns from the lowest quartile appear to be

positive at a low significance, while the returns from the high accruals portfolio remain

insignificant. As the returns of the hedge portfolio are still significantly negative, it suggests

that the monthly risk-free interest rate during the period was higher than the raw mean

monthly return in the low accruals portfolio. One reason why the results are different from

Green et al. (2011) could be the difference in investigation periods. As the investigation

48

period begins only a couple of years before the time that Green et al. says the anomaly would

have been arbitraged away (i.e. 2004), the sample of this study is likely too small to establish

any statistical significance. Perhaps if the sample period had been extended to several years

prior to 2004, the results would have been different.

4.2.4 Results from testing the trading strategy after winsorizing outliers

Kraft et al. (2006) argue that a great deal of the abnormal returns in the accruals based trading

strategy in previous research is attributable to outliers. To investigate if the insignificant

results in this study are caused by outliers, 5% of the abnormal returns are winsorized. While

winsorizing returns does help investigate the reliability of the results, it is not practically

implementable for the investor, which should be kept in mind when assessing the results.

Table 4a below presents the results from the calendar time approach using CAPM and Fama-

French risk adjustments to the total assets benchmark.

Table 4a. Total Assets benchmark (winsorized 5%) Abnormal returns using the calendar time approach for the total assets benchmark. The top value represents the time series means of

abnormal return (alpha) and the number below is the p-value corresponding to the t-test.

Quartile Return CAPM Return Fama-French 3 factor

Quartile 1 -0.010 -0.009

0.06* 0.09*

Quartile 2 0.000 0.002

0.86 0.43

Quartile3 0.000 0.001

0.84 0.51

Quartile 4 -0.011 -0.010

0.03** 0.04**

Hedge portfolio (Q1- Q4) -0.020 -0.020 0.00*** 0.00***

* significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level

Based on the outcome in the table 4a above, it may be understood that the returns in the low

and high accruals portfolios become statistically significant from zero when outliers are

winsorized. Furthermore, the abnormal returns that were previously positive are now

negative. This indicates that the positive abnormal returns presented in Table 1a are due to

49

some outliers with extremely positive monthly abnormal returns, but that the majority of the

abnormal returns in the quartile are substantially lower. The average monthly return for the

majority of extreme accrual firms are therefore likely negative in the year following the

portfolio formation. However, the gains from the short position in high accrual firms are not

sufficient to compensate for the negative returns generated by the long position in low accrual

firms. As a result, the monthly hedge portfolio return is negative at a high significance level.

Interestingly, the abnormal returns for the middle quartiles are no longer significantly

different from zero once the outliers are winsorized, suggesting that the positive abnormal

returns in tables 1a and 2a may have been due to certain outliers. However, winsorization is a

way of influencing the data and should, if possible, be avoided. The results in table 1a and 2a

should therefore be considered more accurate portrayals of the empirical returns from the

trading strategy. The same procedure is done for the earnings benchmark portfolios and the

results are presented in Table 4b below:

¨Table 4b. Earnings benchmark (winsorized 5%) Abnormal returns using the calendar time approach for the earnings benchmark. The top value represents the time series means of abnormal

return (alpha) and the number below is the p-value corresponding to the t-test.

Quartile Return CAPM Return Fama-French 3 factor

Quartile 1 -0.010 -0.009 0.04** 0.06* Quartile 2 0.00 0.002 0.97 0.36 Quartile3 0.001 0.003 0.71 0.20 Quartile 4 -0.012 -0.011 0.02** 0.03** Hedge portfolio (Q1- Q4) -0.019 -0.020

0.00*** 0.00***

* significant at the 10% level, ** significant at the 5% level, *** significant at the 1% level

50

The findings presented in table 4b are similar to the findings in table 4a. The monthly

calendar time abnormal returns for extreme accrual firms are significantly negative, as is the

return for the hedge portfolio.

The findings in tables 4a and 4b do not support the earnings fixation hypothesis proposed by

Sloan (1996) as the earnings fixation hypothesis does not explain the negative abnormal

return earned by the lowest accrual portfolio. On the contrary, the earnings fixation

hypothesis suggests that the earnings surprise from firms in low accrual firms should be

positive, as there are low or few reversals in the year after the accruals have been measured.

The positive earnings surprise should thereby generate positive abnormal returns to the

investor in a low accrual firm. The findings cannot either be explained by the agency theory

of overvalued equity as proposed by Kothari et al. (2006). This theory predicts that the high

accrual quartile would experience negative abnormal returns as the high accruals of

overvalued firms reverse, while the returns of the lowest accrual quartile are likely to be

statistically insignificant. As the results in tables 4a and 4b show that the returns of the low

accruals portfolio (quartile 1) are significantly negative, the findings do not support the

agency theory of overvalued equity.

Instead, the results from the winsorized regressions seem to be in line with the inverted U-

shape pattern in Kraft et al.’s. (2006) study. As mentioned in chapter 2, the inverted U-shape

refers to the relationship between the size of a firm’s accruals and its abnormal stock return. It

suggests that any firm with extreme accruals, either high or low, will likely suffer from

negative abnormal stock returns in the year after the portfolio formation.

According to Kraft et al. (2006) the significantly negative abnormal stock return in the 1st and

4th quartiles suggest that investors overreact to the positive accruals announcement but

underreact to negative accruals. In this explanation, the investors understand the difference in

components but under- and overreact to the information. Intuitively it can be explained as,

investors in low accrual firms should react positively as the firms have relatively high cash

flows, and these have higher persistence in earnings than accruals. However, the majority of

investors underreact to the potential for persistent future earnings, resulting in negative

returns. In the meantime, investors in high accrual firms should react negatively to the high

accruals in their firms’ reporting, as these are less persistent in future earning as cash flows.

51

However, investors in high accrual firms seem to overreact, reducing the stock price too

much, which leads to negative abnormal returns in the high accrual portfolio. As this is not

what alternative theories (such as the earnings fixation hypothesis and the agency theory of

overvalued equity) predict, Kraft et al. (2006) present an alternative explanation; that there

are certain unspecified firms characteristics that are found in extreme accrual firms and may

be associated with the negative returns.

When investigating the characteristics of the firms classified in each portfolio in section 4.1,

there is a pattern present in the total assets benchmark that is not found to the same extent in

the earnings benchmark. In the total assets benchmark, the extreme accruals portfolios are

characterized by smaller firms (measured both as market capitalization and total assets),

substantially lower cash flows from operations and lower stock prices (see figures 4; 8; 5; 7

respectively).

Kraft et al. (2006) argue that low stock prices are an indication that firms have performed

poorly in the past. Assuming this argument is true, the poor past performance may be an

explanation for the expected continued poor performance in extreme accrual firms (sorted on

total assets) that results in negative abnormal returns. As the same pattern is only found to a

limited extent in the earnings benchmark, the results for the earnings benchmark cannot be

explained by the same reasoning.

52

5. CONCLUSION

The final chapter presents conclusions based on the results of the study. The aim is to connect

back to the research questions and suggest possible topics for further research.

The study has investigated the possibility to earn abnormal returns for portfolios based on

firms’ accruals on the Swedish stock market. Each year, between the years 2002 to 2012,

firms listed on NASDAQ OMX Stockholm main market are classified into quartiles based on

the size of their accruals in the previous year’s financial statements. Starting in June each

year, the stock prices are recorded on a monthly basis to calculate the monthly returns on each

stock. The monthly portfolio returns for each quartile and the hedge portfolio are calculated

and then adjusted using three different methods of risk adjustment in order to investigate if

the portfolios yield abnormal returns to an investor.

The first research question in this study was; to what extent is the measure that normalizes

accruals by earnings more effective than normalizing by total assets when attempting to create

a trading strategy exploiting the accruals anomaly? The results show that the benchmarks

differ primarily in relation to the firms classified in the low accruals quartile. The discussion

in the first part of the analysis finds that when comparing the benchmarks’ low accruals

classifications, the earnings benchmark seems to select firms that are larger, have a higher

stock prices (that by extension are argued to have lower relative transaction costs), and higher

cash flows from operations. Therefore, the earnings benchmark seems to be preferable in this

comparison. Unfortunately, the improved classification does not spill over to the abnormal

returns earned by the investor. With relation to abnormal returns, the benchmarks seem to

generate similar results in the hedging strategy, as well as in the individual quartile portfolios.

The second research question in this study was; to what extent is it possible to earn abnormal

returns from an accruals based trading strategy in Sweden? The results do not support that

there may be positive abnormal returns from the accruals based trading strategy in Sweden.

These results are robust to three different methods of estimating abnormal returns. While the

findings initially appear to contradict several previous research papers, the results are not

surprising. Firstly, Green et al. (2011) argue that the anomaly has been arbitraged away due to

the attention the subject has received during the years. While this study fails to find

53

differences in abnormal returns between sub periods, extending the investigation period may

lead to findings that support this view. Furthermore, Kraft et al. (2006) argue that the majority

of previous accruals research is affected by biases, and once these are corrected for the,

results are significantly different. The two main biases are the look-ahead bias and the bias

arising from outliers. By including firms that have been delisted during the portfolio year, the

results in this study are less likely to suffer from an upward bias. Furthermore, after the

outliers have been winsorized, the abnormal returns on the high and low accruals portfolios

are negative and significant. This supports the inverted U-shaped relationship found in Kraft

et al. (2006) that suggests that extreme accrual firms experience negative abnormal returns

during the portfolio year.

From an investor perspective the results suggest that it is unlikely to earn arbitrage profits

from taking long and short positions in firms based on their accruals. It is likely that the

market has already recognized the importance of earnings components and priced firms

accordingly. Thus, in order to make profitable investments, investors should look to other

factors that create value in a firm. From the firm perspective these results suggest that high

accruals need not necessarily be negatively perceived by the market. On the other hand, as

they do not either lead to significantly positive returns, firms should be careful not to use

doubtful accrual criteria in an attempt to manipulate and increase earnings.

As the results are not in line with the earnings fixation hypothesis, or the agency theory of

overvalued equity, there is a need for a different theoretical explanation for the inverted U-

shaped pattern. While this study has not been able to find any generally applicable firm

characteristics that may explain the results, future research may find this to be an interesting

task. Future research may also choose to focus on differences in accruals between industries.

Perhaps the anomaly is more likely to exist within certain industries? However, in order to do

this, it is recommended to choose a broader market than the one in this study in order to

ensure enough observations within each industry group.

54

REFERENCES

Academic literature

Ball, R. & Brown, P. (1968) “An empirical evaluation of accounting income numbers” Journal of

Accounting Research, p. 159-178.

Barber, B. & Lyon, J. (1997) ”Detecting long-run abnormal stock returns: the empirical power and

specification of test statistics” Journal of Financial Economics, vol. 43, p. 341-373.

Barber, B. Lyon, J. & Tsai, C. (1999) “Improved Methods for Tests of Long-run Abnormal Stock

Returns” Journal of Finance, vol.54, no.1, p. 165–201.

Bernard, V.L. & Thomas, J.K. (1990) “Evidence that stock prices do not fully reflect the implications

of current earnings for future earnings” Journal of Economics & Accounting, vol. 13, p. 305-340.

Bradshaw, M.T. Richardson, S.A. & Sloan, R.G. (2001) "Do analysts and auditors use information in

accruals?" Journal of Accounting Research, vol. 39, no. 1, p. 45-74.

Collins, D. & Hribar, P. (2000) “Earnings-based and accruals-based market anomalies: one effect or

two?” Journal of Accounting & Economics, vol. 29, no. 1, p. 101-123.

Dechow, P. Richardson, S. & Sloan, R. (2005) “The Persistence and Pricing of the Cash Component

of Earnings” Journal of Accounting Research, vol. 46, no. 3, p. 537-566.

Fama, E.F. (1970)”Efficient Capital Markets: A Review of Theory and Empirical Work” Journal of

Finance. vol. 25, no. 2, p. 383-417.

Fama, E. F. (1998) ”Market efficiency, long-term returns and behavioral finance” Journal of Financial

Economics, vol. 49, p. 283-306.

Fama, E.F. & French K.R. (1992) “The Cross-Section of Expected Stock Returns” The Journal of

Finance, vol. 67, no. 3, p. 427-465.

Fama, E.F. & French, K.R. (1993) "Common risk factors in the returns on stocks and bonds" Journal

of Financial Economics, vol. 33, no. 1, p. 3-56.

Green, J. Hand, J.R.M. & Soliman, M.T. (2011) "Going, Going, Gone? The apparent demise of the

accruals anomaly” Management Science, vol. 57, no. 5, p. 797-816.

Hafzalla, N. Lundholm, R. J. & Van Winkle, M. (2007) “Repairing the Accruals Anomaly” Working

paper. Available from: http://ssrn.com/abstract=1030855. [Accessed 24 May 2014]

Hand, J. (1990) “A test of the extended functional fixation hypothesis” The Accounting Review, vol.

65, no. 4, p. 740-763.

Hribar, P. & Collins, D.W (2002) “Errors in Estimating Accruals: Implications for Empirical

Research” Journal of Accounting Research, vol 40, no 1, p. 105–134.

Hung, M. (2001) “Accounting standards and value relevance of financial statements: An international

analysis” Journal of Accounting and Economics, vol. 30, no. 3, p. 401–420.

55

Jensen, M.C. (2005) “Agency Costs of Overvalued Equity” Financial Management, vol.34, no. 1, p.

5-19.

Johnson, N. J. (1978) “Modified t tests and confidence intervals for asymmetrical population” Journal

of the American Statistical Association, vol. 73, p. 536–544.

Johnson, B. & Schwartz, W. (2001) “Evidence that capital markets learn from academic research:

Earnings surprise and the persistence of post-earnings announcement drift” Investor Relations

Quarterly, vol, 4.

Khan, M. (2008) “Are accruals mispriced? Evidence from tests of an intertemporal capital asset

pricing model” Journal of Accounting and Economics, vol. 45, no.1, p. 55-77.

Kothari, S. Sabino, J. & Zach, T. (2005) “Implications of Survival and Data Trimming for Tests of

Market Efficiency.” Journal of Accounting & Economics, vol. 39, p.129–61.

Kothari, S.P. Loutskina, E. & Nikolaev, V.V (2006) “Agency Theory of Overvalued Equity as an

Explanation for the Accrual Anomaly” Working paper. Available from:

http://ssrn.com/abstract=871750 [Accessed 24 May 2014]

Kraft, A. Leone, A.J. & Wasley, C. (2006) “An analysis of the theories and explanations offered for

the mispricing of accruals and accrual components” Journal of Accounting Research, vol. 44, no. 2, p.

297-339.

LaFond, R. (2005) “Is the Accrual Anomaly a Global Anomaly?” MIT Sloan Research Paper No.

4555-05. Available from: http://ssrn.com/abstract=782726 [Acessed 24 May 2014]

La Porta, R. Lopez-de-Silanes, F. Shleifer, A. & Vishny, R. (1998) “Law and finance” Journal of

Political Economy, vol. 106, no. 6, p. 1113–1155.

Lintner, J. (1965) "The valuation of risk assets and the selection of risky investments in stock

portfolios and capital budgets" The Review of Economics and Statistics, vol. 47, no. 1, p. 13-

37

Loughran, T. & Ritter, J. (2000) “Uniformly least powerful tests of market efficiency” Journal of

Financial Economics, vol. 55, p. 361-389.

Maines, L. & Hand, J. (1996) “Individuals’ perceptions and misperceptions of the time series

properties of quarterly earnings” The Accounting Review, vol. 71, no. 3, p. 317-336.

Mitchell, M. L. & Stafford, E. (2000) “Managerial Decisions and Long-term Stock Price

Performance” Journal of Business, vol. 73. no 1, p. 287-320.

Ou, J. & Penman, S. (1989) “Financial statement analysis and the prediction of stock returns” Journal

of Accounting and Economics, vol. 11, no. 4, p. 295-330.

Pincus, M. Rajgopal, S. & Venkatachalam, M. (2007) "The Accrual Anomaly: International Evidence"

The Accounting Review, vol. 82, no. 1, p. 169-203.

Sharpe, W.F. (1964) "Capital asset prices: A theory of market equilibrium under conditions of risk"

Journal of Finance, vol. 19, no. 3, p. 425-442.

56

Sloan, R. (1996) “Do stock prices fully reflect information in accruals and cash flows about future

earnings?” Accounting Review, vol. 71, no. 3, p. 289-315.

Wu, J. Zhang, L. & Zhang, X.F. (2009) “The q-theory approach to understanding the accruals

anomaly” Journal of Accounting Research, vol. 48, no.1, p. 177-223.

Xie, H. (2001) "The Mispricing of Abnormal Accruals" Accounting Review, vol. 76, no. 3, p. 357-

373.

Zach, T. (2003) “Inside the Accrual Anomaly” Working paper. Available from:

1.196.3925&rep=rep1&type=pdf [Accessed 24 May 2014]

Books

Bernstein, L. (1993) Financial Statement Analysis. 5th edition. Homewood, IL: Irwin.

Brooks, C. (2008) Introductory Econometrics for Finance. 2nd

edition. Cambridge: University Press.

Shefrin, H. (2007) Behavioral Corporate Finance. Decisions that Create Value. New York,

NY:McGraw- Hill/Irwin.

Student work

Giaever, H. & Gabrielsson, T. (2007) “The Accruals Anomaly in Sweden” Master Thesis in Finance,

Lund School of Economics & Management.

Ivanov, B. & Maximova, V. (2011) “Are accruals mispriced? Evidence from Sweden” Master Thesis

in Accounting and Financial Management, Stockholm School of Economics.

Additional sources

Henry, D. (2004) “Fuzzy Numbers” Business week, 4 October 2004, p. 79-88 [Online]. Available

from: http://www.businessweek.com/stories/2004-10-03/fuzzy-numbers. [Accessed 24 May 2014]

Nobel Foundation (2013) The Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred

Nobel 2013. [Press release]. Available from: http://www.nobelprize.org/nobel_prizes/economic-

sciences/laureates/2013/press.html [Accessed 24 May 2014].

Skypala, P. (2009) “'Quant creator' says now is a good time to be different” Financial Times, 26

January 2009, p. 4 [Online] Available from: http://www.ft.com/intl/cms/s/0/a3a49ce2-eb4a-11dd-

bb6e-0000779fd2ac.html#axzz32dYCqPor. [Accessed 24 May 2014]

57

APPENDIX

Appendix 1

Figure 1. Summary of test statistics calculations when testing the buy-and-hold

abnormal returns:

ATA Q1 ATA Q2 ATA Q3 ATA Q4 NI Q1 NI Q2 NI Q3 NI Q4 NI HEDGE ATA HEDGE

2002 -0,0356 0,0241 0,0098 0,0032 -0,0594 0,0505 0,0265 -0,0150 -0,0444 -0,0388

2003 0,0270 0,0365 -0,0812 0,0146 0,2261 -0,0379 -0,1914 -0,0013 0,2274 0,0124

2004 0,0585 -0,0344 0,1006 -0,1254 0,0007 0,1264 0,0219 -0,1465 0,1471 0,1839

2005 0,0551 -0,0612 0,0484 -0,0435 -0,0093 0,1021 -0,0468 -0,0440 0,0348 0,0986

2006 -0,0738 0,0766 0,0081 -0,0063 -0,0105 -0,0161 0,0125 0,0142 -0,0246 -0,0675

2007 -0,0441 -0,0038 0,0251 0,0232 -0,0224 -0,0344 0,0316 0,0251 -0,0475 -0,0673

2008 -0,0365 -0,0230 0,0054 0,0250 -0,0151 -0,0372 0,0444 -0,0208 0,0057 -0,0614

2009 0,0019 -0,0273 -0,0241 0,0495 -0,0871 -0,0249 0,0531 0,0590 -0,1461 -0,0476

2010 0,0647 -0,0069 -0,0345 -0,0243 0,0550 0,0414 -0,0781 -0,0196 0,0745 0,0889

2011 -0,0417 -0,0131 0,0302 0,0249 -0,0504 -0,0065 0,0456 0,0121 -0,0625 -0,0665

2012 -0,0436 0,0253 -0,0483 0,0659 0,0481 -0,1017 -0,0007 0,0507 -0,0027 -0,1095

n 11 11 11 11 11 11 11 11 11 11

Average portfolio BHAR -0,0062 -0,0007 0,0036 0,0006 0,0069 0,0056 -0,0074 -0,0078 0,0147 -0,0068

Standard Deviation 0,0496 0,0385 0,0498 0,0520 0,0840 0,0675 0,0732 0,0555 0,1039 0,0917

Conventional t-stat -0,4139 -0,0580 0,2386 0,0398 0,2719 0,2748 -0,3359 -0,4677 0,4697 -0,2463

S -0,1248 -0,0175 0,0720 0,0120 0,0820 0,0829 -0,1013 -0,1410 0,1416 -0,0742

γ 0,2839 0,4003 0,1522 -1,0766 1,4378 0,3831 -1,3944 -1,1509 0,5711 0,8229

Skewness adj t-stat -0,3947 -0,0378 0,2472 -0,0145 0,3549 0,2970 -0,4218 -0,5508 0,5111 -0,1999

Average BHAR each year

58

Appendix 2

OLS corrections

For several of the calendar time regressions, the Jarque-Bera null hypothesis is rejected.

However, as the regression for each portfolio consists of at least 581 observations, the non-

normality is assumed to not have a significant impact. The tests for the remaining

assumptions all show that the regressions fulfil the OLS requirements, those that are corrected

for are presented in figure 2 and 3 below:

Figure 2. Corrections in Percent Accruals calendar time regressions:

Figure 3. Corrections in Size Accruals calendar time regressions:

Quartile CAPM Fama-French 3 factor

Quartile 1 none Whites robust SE

Quartile 2 none Whites robust SE

Quartile 3 none Whites robust SE

Quartile 4 none Whites robust SE

Hedge portfolio Whites robust SE Whites robust SE

Quartile CAPM Fama-French 3 factor

Quartile 1 none none

Quartile 2 none Whites robust SE

Quartile 3 none Whites robust SE

Quartile 4 none none

Hedge portfolio Whites robust SE Whites robust SE

59

Appendix 3.

Figure 4b

Figure 5b

Figure 6b

Figure 7b

Mean Median Mean Median

Quartile 1 5310 459 9470 864

Quartile 2 18167 1205 12200 894

Quartile 3 11671 1310 12518 1040

Quartile 4 6037 995 6856 920

(mSEK)

Size accruals Percent Accruals

Market capitalization

Mean Median Mean Median

Quartile 1 80,50 26,47 78,58 34,17

Quartile 2 105,86 42,40 105,73 39,85

Quartile 3 67,11 44,40 70,92 43,67

Quartile 4 80,54 33,69 78,89 32,94

(SEK)

Size Accruals Percent Accruals

Stock Price

Mean Median Mean Median

Quartile 1 88 -9 177 11

Quartile 2 811 50 679 45

Quartile 3 756 73 763 57

Quartile 4 509 55 542 53

(mSEK)

Size accruals Percent Accruals

Net income before extraordinary events

Mean Median Mean Median

Quartile 1 671 48 1277 106

Quartile 2 1831 127 1283 85

Quartile 3 1057 105 950 65

Quartile 4 216 17 248 12

(mSEK)

Size accruals Percent Accruals

Cash flow from operations

60

Figure 8a

Figure 8b

Mean Median Mean Median

Quartile 1 4926 513 13667 1299

Quartile 2 17508 1486 12287 1124

Quartile 3 13740 1618 9556 816

Quartile 4 6611 833 7049 786

(mSEK)

Size accruals Percent Accruals

Total assets